An Investigation of the Structural Characteristics of the Indian IT Sector and the Capital Goods Sector: An Application of the R Programming in Time Series Decomposition and Forecasting
JJournal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 *Corresponding Author: [email protected] Journal of Insurance and Financial Management (ISSN: 2371-2112) All rights reserved
An Investigation of the Structural Characteristics of the Indian IT Sector and the Capital Goods Sector – An Application of the R Programming in Time Series Decomposition and Forecasting
Jaydip Sen a,* , Tamal Datta Chaudhuri b a,b Calcutta Business School, Bishnupur – 743503, West Bengal, India
ABSTRACT
Time series analysis and forecasting of stock market prices has been a very active area of research over the last two decades. Availability of extremely fast and parallel architecture of computing and sophisticated algorithms has made it possible to extract, store, process and analyze high volume stock market time series data very efficiently. In this paper, we have used time series data of the two sectors of the Indian economy – Information Technology (IT) and Capital Goods (CG) for the period January 2009 – April 2016 and have studied the relationships of these two time series with the time series of DJIA index, NIFTY index and the US Dollar to Indian Rupee exchange rate. We establish by graphical and statistical tests that while the IT sector of India has a strong association with DJIA index and the Dollar to Rupee exchange rate, the Indian CG sector exhibits a strong association with the NIFTY index. We contend that these observations corroborate our hypotheses that the Indian IT sector is strongly coupled with the world economy whereas the CG sector of India reflects India’s internal economic growth. We also present several models of regression between the time series which exhibit strong association among them. The effectiveness of these models have been demonstrated by very low values of their forecasting errors.
Journal of Insurance and Financial Management All rights reserved
ARTICLE INFO
JEL Classification:
G11 G14 G17 C63
Keywords:
Time Series Decomposition Trend Seasonal Random R Programming Language Association Tests Cross Correlation Linear Models Correlation Regression . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 Introduction
The literature on portfolio choice and forecasting of stock returns has concentrated on various characteristics of companies like their Profitability Indicators, Leverage, P/E ratio, P/BV ratio, Size, Volume of Trade, Market Capitalization and Dividend Payout Ratio. Understanding companies by the above mentioned parameters, somehow, leads to some standardization and robs the companies of their individuality. Every company is distinct, and one of the sources of this distinctiveness stems from the sector to which it belongs. Each sector is tied to some aspects of the economy. These aspects may be socio economic and/or demographic characteristics, the income distribution pattern, the extent of global integration, domestic endowment of resources, and state of the technology or market size. The literature has not explicitly captured these aspects, and hence sectoral distinctions have not been modelled adequately. As a consequence, any methodology for forecasting of stock returns for a sample set of diverse companies, we feel, falls short of the desired results. In our previous work, we have been emphasizing on this specific aspect of sectoral characteristics (Sen and Datta Chaudhuri, 2016a; Sen and Datta Chaudhuri, 2016b; Sen and Datta Chaudhuri, 2016c). Following our decomposition approach, we have demonstrated that indeed the sectors are different in terms of their trend, seasonal and random components. We have pointed out that, for India, each of the above components is tied to some social or economic feature, and the forecasting methodology that we suggested in our work incorporates these sectoral characteristics. In this paper, we look at two different sectors in India, namely the Information Technology (IT) and the Capital Goods (CG) sectors, and demonstrate that these two sectors are completely different in terms of their behavioral characteristics. We hypothesize that while the IT sector, being a services sector, is tied to the rest of the world, the CG sector is very much tied to the India story. We use the R programming framework to decompose the time series of the IT and CG sectors stock market index into trend, seasonal and random components, and then relate the movement of each component to the components of Dow Jones Industrial Average (DJIA), NIFTY (Indian National Stock Exchange Index) and the US Dollar to Indian Rupee Exchange Rate. Our contention is that instead of comparing the movement of the aggregates, one should compare the movement of the components for better understanding of the sectors. This would J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 give further insight into the choice of stocks for portfolio formation and also in portfolio redesigning. The contribution of this work is threefold. First, we propose a time series decomposition approach and then illustrate that the proposed technique provides us with a deeper understanding of the behavior of a time series by observing the relative magnitudes of its three components namely trend, seasonal and random. Second, we present mechanisms of studying associations among different time series using various graphical and statistical tests. The association analysis provides us further insights into the behavioral characteristics of different time series. Several hypotheses are also validated using the association analysis. Third, we develop various models of regression for time series that exhibit strong associations among them in our study. The models are constructed using suitably designed training data sets and then tested using appropriate test data sets. The forecast accuracies of each of the models are computed so as to have an idea about their efficacies and robustness. The rest of the paper is organized as follows. Section 2 briefly discusses the methodology in constructing various time series and decomposing the time series into its components. It also presents a brief outline on the forecasting frameworks designed in this work using the R programming language. Section 3 provides a detailed discussion on the methods of decomposition, the decomposition results of all the sectors under study, and an analysis of the results. Section 4 presents a methodology of comparing and analyzing association between several time series under our investigation. The association between the Indian IT sector and Indian CG sector with DJIA index, US Dollar to Indian Rupee exchange rate, and the NIFTY index are studied in great detail. We present several hypotheses and validate them through graphical means and several statistical tests. Section 5 presents several linear models for forecasting that enables one to forecast the index of one sector given the index of another sector to which is known to be strongly associated. We present eight models and present extensive results to demonstrate their efficacy and effectiveness in forecasting. In Section 6, we discuss some related work in the current literature. Finally, Section 7 concludes the paper. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 Methodology
In this section, we provide a brief outline of the methodology that we have followed in our work. However, each of the following sections contains detailed discussion on the methodology followed in the work related to that Section. We have used the
R programming language (Ihaka & Gentleman, 1996) for data management, data analysis and presentation of results. R is an open source language with very rich libraries that is ideally suited for data analysis work. We use daily data of the Indian IT sector index, Indian CG sector index, NIFTY index, DJIA index and the US Dollar to Indian Rupee exchange rate for the period January 2009 to April 2016. The daily index values are first stored in five plain text files – each sector data in one file. The daily data are then aggregated into monthly averages resulting in 88 values in the time series data. These 88 monthly average values for each sector are stored in five separate plain text files – each sector monthly average in one file. The records in the text file for each sector are read into an R variable using the scan( ) function in R. The resultant R variable is converted into a monthly time series variable using the ts( ) function defined in the
TTR library in the R programming language. The monthly time series variable in R is now an aggregate of its three constituent components: (i) trend, (ii) seasonal, and (iii) random. We then decompose the time series into its three components. For this purpose, we use the decompose( ) function defined in the TTR library in R. The decomposition results enable us to make a comparative analysis of the behavior of the five time series belonging to five different sectors. We validate several hypotheses by our deeper analysis of the decomposition results. After a detailed analysis of the decomposition results, we enter into our second endeavor in this work. Based on our deeper understanding about the association among different sectors as observed from their time series analysis, we make bivariate analysis and forecasting using linear regression models. This analysis enables us to forecast the performance of one sector on basis of performance of another to which it is closely associated with. We have also carried out analysis on the forecast accuracies by suitably choosing our training data set for building the linear regression model and test data for testing the effectiveness of our forecasting models. In our previous work, we have highlighted the effectiveness of time series decomposition approach for robust analysis and forecasting of the Indian Auto sector (Sen & Datta Chaudhuri, 2016a; Sen & Datta Chaudhuri, 2016b) and we have also made a comparative study of the behavioral characteristics of two different sectors of the Indian economy – the Consumer J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Durable Goods sector and the Small Cap sector (Sen & Datta Chaudhuri, 2006c). In contrast to our previous work, in this paper, we have presented a detailed study on the structural decomposition of the time series index of Indian IT and CG sectors, NIFTY index, DJIA index and the US Dollar to the Indian Rupee exchange rate. We have then carried out association analyses between these time series to validate several hypotheses that we postulate. Association analyses are carried out using several robust statistical tests. After a comprehensive association analysis, we have constructed several linear regression models between those time series which exhibited strong association among them. To demonstrate the robustness and accuracies of the linear models, we have used the models for forecasting using suitable chosen training and test data sets. Time Series Decomposition Results
We now present the methods that we have followed to decompose the time series of five different sectors – Indian IT sector, Indian CG sector, DJIA index, NIFTY index and US Dollar to Indian Rupee exchange rate. For all these sectors, we have first taken the daily index values from January 2009 to April 2016 and saved them in five separate plain text (.txt) files – one file storing the daily time series index of one sector. From these daily index values, we have computed the monthly averages and saved the monthly average values in five separate text files. Each of these text files contained 88 values (records of 7 years and 4 month leading to 88 monthly average values). We used R language function scan ( ) to read these text files and store them into five appropriate R variables. Then, we converted these five R variables into five time series variables using the R function ts ( ), which is defined in the package
TTR . Once these five time series variables are constructed, we have used the plot ( ) function in R to derive the displays of the time series for the five sectors under study for the period January 2009 – April 2016. The time series for the Indian IT sector index, the Indian CG sector index, the DJIA index, the NIFTY index and the US Dollar to Indian Rupee exchange rate values are represented in Figure 1, Figure 2, Figure 3, Figure 4, and Figure 5 respectively. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 Figure 1
The Indian IT sector index time series (Jan 2009 – Apr 2016)
Figure 2
The Indian CG sector index time series (Jan 2009 – Apr 2016)
Figure 3
The DJIA index time series (Jan 2009 – Apr 2016) J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Figure 4
The NIFTY index time series (Jan 2009 – Apr 2016) The plots of the time series for the five sectors exhibit the overall behavior of these time series over the period under consideration (i.e., January 2009 – December 2016). However, to get a deeper insight into these time series, we have decomposed the five time series variables into their trend, seasonal and random components using the decompose ( ) function that is defined in the TTR library in the R programming environment. Figure 6, Figure 7, Figure 8, Figure 9 and Figure 10 depict the decomposition results for the time series of the Indian IT sector, Indian CG sector, the DJIA index, the NIFTY index and the US Dollar to Indian Rupee exchange rate values. Each of the five figures (Figure 6 to Figure 10) has four boxes arranged in a stack. The boxes depict the overall time series, the trend, the seasonal and the random component respectively, arranged from top to bottom.
Figure 5
The US Dollar to Indian Rupee exchange rate time series (Jan 2009 – Apr 2016) . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 Figure 6
Decomposition of Indian IT sector index time series into its trend, seasonal and random components (Jan 2009 – Apr 2016)
Figure 7
Decomposition of Indian CG sector index time series into its trend, seasonal and random components (Jan 2009 – Apr 2016)
Figure 8
Decomposition of DJIA index time series into its trend, seasonal and random components (Jan 2009 – Apr 2016) J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Figure 9
Decomposition of the NIFTY index time series into its trend, seasonal and random components (Jan 2009 – Apr 2016)
Figure 10
Decomposition of the US Dollar to Indian Rupee exchange rate time series into its trend, seasonal and random components (Jan 2009 – Apr 2016) The numerical values of the time series and its three components for the Indian IT sector, Indian CG sector, the DJIA index, the NIFTY index, and the US Dollar to Indian Rupee exchange rate values are presented in Table 1, Table 2, Table 3, Table 4 and Table 5 respectively. It may be interesting to observe that the values of the trend and the random components are not available for the period January 2009 – June 2009 and also for the period November 2015 – April 2016. Since the decompose( ) function in R uses a 12 month moving average method for computing the trend component, in order to compute the trend value for January 2009, we need time series data from July 2008 to December 2008. However, since we have used time series data from January 2009 to April 2016, the first trend value the decompose ( ) function could compute was for the month of July 2009 and the last month being November . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 decompose ( ) function first detrends (subtracts the trend component from the overall time series) the time series and arranges the time series values in a 12 column format. The seasonal values for each month is derived by computing the averages of each column. The value of the seasonal component for a given month remains the same for the entire period under study. The random components are obtained after subtracting the sum of the corresponding trend and seasonal components from the overall time series values. Since the trend values for the period January 2009 – June 2009 and November 2015 – April 2016 are missing, the random components for those periods could not be computed as well. Table 1
Aggregate value of the Indian IT sector index time series and its components (Jan 2009 – Apr 2016)
Year Month Aggregate Trend Seasonal Random
January 2189 299 February 2140 367 March 2175 219 April 2483 -181 May 2850 -429 June 3237 -372 July 3500 3550 -175 125 August 4022 3795 -63 290 September 4403 4049 26 328 October 4449 4303 111 35 November 4670 4521 81 68 December 4974 4703 116 155
January 5197 4869 299 29 February 5026 5012 367 -353 March 5381 5132 219 30 April 5379 5258 -181 302 May 5177 5384 -429 222 June 5290 5506 -372 156 July 5423 5628 -175 -30 August 5525 5739 -63 -151 September 5787 5825 26 -64 October 6086 5901 111 74 November 6067 5978 81 8 December 6482 6042 116 324
January 6635 6094 299 242 February 6254 6101 367 -214 March 6207 6054 219 -66 J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
April 6382 6000 -181 563 May 6015 5958 -429 486 June 5984 5909 -372 445 July 5975 5840 -175 310 August 5141 5794 -63 -590 September 5062 5782 26 -746 October 5511 5746 111 -346 November 5637 5699 81 -143 December 5738 5667 116 -45
January 5709 5630 299 -220 February 6089 5626 367 96 March 6065 5682 219 164 April 5676 5731 -181 126 May 5575 5747 -429 257 June 5658 5754 -372 276 July 5425 5784 -175 -184 August 5592 5841 -63 -186 September 5957 5898 26 33 October 5785 5936 111 -262 November 5759 5965 81 -287 December 5768 6005 116 -353
January 6409 6105 299 5 February 6760 6273 367 120 March 6761 6451 219 91 April 5896 6650 -181 -573 May 6039 6875 -429 -407 June 6168 7128 -372 -588 July 7300 7392 -175 83 August 7764 7624 -63 203 September 8048 7820 26 202 October 8473 8023 111 339 November 8466 8247 81 138 December 9133 8483 116 534
January 9379 8713 299 367 February 9373 8918 367 88 March 8843 9122 219 -498 April 8684 9315 -181 -450 May 8621 9513 -429 -463 June 9247 9675 -372 -56 July 9748 9791 -175 132 August 10226 9952 -63 337 September 10494 10167 26 301 October 10650 10384 111 155 . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 November 11061 10571 81 409 December 10414 10718 116 -420
January 10882 10816 299 -233 February 11724 10901 367 456 March 11667 10973 219 475 April 11073 11028 -181 226 May 10721 11053 -429 97 June 10656 11070 -372 -42 July 10703 11087 -175 -209 August 11301 11041 -63 323 September 11164 10970 26 168 October 11296 10952 111 233 November 10999 December 10884
January 10832 299 February 10670 367 March 11023 219 April 11270 -181
Table 2
Aggregate value of the Indian CG sector index and its components (Jan 2009 – Apr 2016)
Year Month Aggregate Trend Seasonal Random 2009
January February March April May June July August September October November December 6588 6144 5906 7576 9823 12597 12208 12540 13315 13698 13234 13714 10918 11514 12138 12746 13173 13391 -309 -507 -213 -47 -284 359 817 41 -6 156 33 -39 474 986 1183 796 28 362
January February March April May June July August September October November December 13926 13111 13927 14140 13497 14163 14850 14712 15411 16126 15919 15296 13566 13767 13945 14133 14346 14524 14596 14594 14536 14467 14417 14353 -309 -507 -213 -47 -284 359 817 41 -6 156 33 -39 669 -149 196 54 -565 -720 -563 78 880 1503 1469 982
January February 14079 12898 14267 14109 -309 -507 121 -704 J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
March April May June July August September October November December 12766 13641 12795 13320 13633 12128 11762 10786 10070 8786 13849 13474 13008 12493 12023 11726 11519 11253 10934 10611 -213 -47 -284 359 817 41 -6 156 33 -39 -869 214 71 468 793 362 249 -623 -897 -1787
January February March April May June July August September October November December 9306 10530 10174 9851 8928 9447 9907 9855 10085 11136 10829 11003 10295 10045 9880 9825 9871 9995 10146 10179 10122 10067 10083 10112 -309 -507 -213 -47 -284 359 817 41 -6 156 33 -39 -679 992 507 73 -659 -907 -1056 -364 -31 913 713 930
January February March April May June July August September October November December 10717 9894 9451 9262 9890 9172 9018 7390 7753 8528 9188 10128 10063 9923 9723 9518 9341 9236 9156 9102 9174 9378 9657 10092 -309 -507 -213 -47 -284 359 817 41 -6 156 33 -39 963 477 -59 -208 834 -423 -954 -1753 -1415 -1006 -502 74
January February March April May June July August September October November December 9669 9661 11400 12225 13603 15918 15782 14577 15048 14475 15810 16149 10655 11237 11840 12392 12915 13442 13965 14545 15107 15571 15897 16059 -309 -507 -213 -47 -284 359 817 41 -6 156 33 -39 -677 -1069 -226 -119 972 2117 1001 -8 -53 -1252 -121 129
January February March April May June July August 16189 17064 17495 17266 16389 17013 18206 17471 16206 16427 16566 16637 16635 16497 16278 15925 -309 -507 -213 -47 -284 359 817 41 293 1143 1142 677 38 157 1112 1506 . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 September October November December 15485 15734 14501 14155 15497 15102 -6 156 33 -39 -7 476
January February March April 12914 11875 12424 12837 -309 -507 -213 -47
Table 3
Aggregate value of the DJIA index and its components (Jan 2009 – Apr 2016)
Year Month Aggregate Trend Seasonal Random 2009
January 8396 48 February 7690 203 March 7202 239 April 7914 343 May 8302 273 June 8581 -19 July 8516 8897 -85 -296 August 9275 9090 -247 432 September 9584 9335 -417 665 October 9802 9606 -255 452 November 10033 9837 -107 303 December 10412 10006 23 383
January 10516 10139 48 330 February 10186 10254 203 -272 March 10606 10338 239 29 April 10995 10422 343 230 May 10769 10520 273 -24 June 10170 10609 -19 -420 July 10116 10702 -85 -500 August 10454 10838 -247 -138 September 10411 10983 -417 -156 October 10987 11103 -255 139 November 11207 11238 -107 76 December 11359 11395 23 -59
January 11802 11575 48 180 February 12183 11710 203 270 March 12084 11779 239 66 April 12396 11828 343 224 May 12599 11874 273 452 June 12104 11927 -19 196 July 12508 11981 -85 612 August 11302 12043 -247 -494 September 11228 12116 -417 -472 J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
October 11349 12185 -255 -581 November 11932 12213 -107 -174 December 11903 12235 23 -355
January 12565 12270 48 248 February 12916 12360 203 353 March 13102 12530 239 333 April 13035 12705 343 -14 May 12615 12828 273 -486 June 12621 12925 -19 -285 July 12829 13036 -85 -122 August 13152 13146 -247 253 September 13446 13265 -417 598 October 13349 13406 -255 198 November 12873 13593 -107 -613 December 13279 13806 23 -550
January 13864 14018 48 -201 February 14251 14207 203 -159 March 14622 14361 239 22 April 14912 14530 343 38 May 15219 14756 273 190 June 15128 15015 -19 131 July 15410 15231 -85 265 August 15108 15401 -247 -46 September 15181 15558 -417 40 October 15679 15697 -255 237 November 15961 15828 -107 240 December 16415 15967 23 424
January 15902 16099 48 -245 February 16289 16236 203 -150 March 16350 16387 239 -276 April 16531 16509 343 -321 May 16737 16632 273 -167 June 16958 16765 -19 212 July 16743 16894 -85 -66 August 17066 17034 -247 279 September 16852 17169 -417 99 October 16918 17294 -255 -121 November 17674 17413 -107 368 December 17893 17511 23 358
January 17534 17596 48 -110 February 18005 17642 203 160 March 17891 17624 239 28 April 17992 17605 343 44 . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 May 18116 17607 273 236 June 17945 17597 -19 366 July 17792 17536 -85 342 August 17117 17416 -247 -52 September 16367 17324 -417 -540 October 16944 17287 -255 -88 November 17697 -107
December 17639 January 16312 48 February 16348 203 March 17344 239 April 17665 343
Table 4
Aggregate value of the NIFTY index and its components (Jan 2009 – Apr 2016)
Year Month Aggregate Trend Seasonal Random 2009
January 2854 -5 February 2819 -42 March 2802 -17 April 3360 50 May 3958 -67 June 4436 -99 July 4343 4183 35 124 August 4571 4364 -53 261 September 4859 4547 -32 344 October 4994 4726 143 125 November 4954 4853 81 20 December 5100 4930 6 164
January 5156 5003 -5 158 February 4840 5083 -42 -200 March 5178 5159 -17 36 April 5295 5245 50 1 May 5053 5337 -67 -217 June 5188 5419 -99 -132 July 5360 5481 35 -156 August 5457 5531 -53 -20 September 5811 5569 -32 274 October 6096 5607 143 346 November 6055 5648 81 326 December 5971 5678 6 287
January 5783 5700 -5 89 February 5401 5694 -42 -250 J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
March 5538 5645 -17 -89 April 5839 5568 50 221 May 5492 5481 -67 77 June 5473 5388 -99 184 July 5597 5303 35 259 August 5077 5267 -53 -137 September 5016 5257 -32 -210 October 5060 5223 143 -306 November 5004 5177 81 -254 December 4782 5138 6 -363
January 4920 5106 -5 -181 February 5409 5101 -42 350 March 5298 5131 -17 184 April 5254 5177 50 28 May 4967 5231 -67 -197 June 5074 5307 -99 -134 July 5222 5400 35 -213 August 5330 5464 -53 -81 September 5485 5498 -32 19 October 5689 5540 143 6 November 5680 5609 81 -10 December 5932 5682 6 244
January 6008 5736 -5 277 February 5846 5770 -42 119 March 5672 5796 -17 -107 April 5891 5835 50 7 May 5997 5875 -67 189 June 5779 5908 -99 -29 July 5827 5925 35 -133 August 5528 5951 -53 -370 September 5925 6016 -32 -59 October 6171 6100 143 -72 November 6169 6203 81 -115 December 6223 6339 6 -122
January 6124 6494 -5 -365 February 6369 6678 -42 -266 March 6695 6866 -17 -154 April 6899 7038 50 -188 May 7450 7222 -67 294 June 7591 7403 -99 287 July 7739 7587 35 117 August 8025 7786 -53 292 September 7944 7967 -32 8 . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 October 8276 8117 143 16 November 8497 8220 81 196 December 8241 8281 6 -46
January 8518 8337 -5 186 February 8750 8381 -42 412 March 8664 8388 -17 293 April 8524 8379 50 96 May 8300 8350 -67 17 June 8196 8308 -99 -13 July 8477 8249 35 193 August 8337 8144 -53 246 September 7816 8033 -32 -185 October 8169 7949 143 77 November 7913 December 7818 January 7536 -5 February 7200 -42 March 7550 -17 April 7632 50
Table 5
Aggregate value of the US Dollar to Indian Rupees exchange rate time series and its components (Jan 2009 – Apr 2016)
Year Month Aggregate Trend Seasonal Random 2009
January 49 -0.2 February 49 -0.6 March 51 -0.7 April 50 -1.4 May 50 -0.4 June 47 0.4 July 48 48.3 0.2 -0.4 August 48 48 1.1 -1.1 September 49 47.7 0.6 0.7 October 47 47.3 -0.2 -0.1 November 47 46.9 0.6 -0.4 December 46 46.7 0.7 -1.3
January 46 46.6 -0.2 -0.4 February 46 46.5 -0.6 0.1 March 46 46.4 -0.7 0.3 April 45 46.1 -1.4 0.3 May 45 46 -0.4 -0.6 June 47 45.8 0.4 0.8 July 47 45.8 0.2 1.1 J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
August 47 45.8 1.1 0.2 September 46 45.7 0.6 -0.3 October 44 45.6 -0.2 -1.5 November 45 45.6 0.6 -1.1 December 45 45.5 0.7 -1.2
January 46 45.3 -0.2 0.9 February 46 45.1 -0.6 1.5 March 45 45.2 -0.7 0.5 April 44 45.5 -1.4 -0.1 May 45 46 -0.4 -0.7 June 45 46.7 0.4 -2 July 44 47.2 0.2 -3.4 August 46 47.5 1.1 -2.6 September 49 47.9 0.6 0.5 October 49 48.5 -0.2 0.6 November 52 49.3 0.6 2.1 December 53 50.2 0.7 2.1
January 51 51.2 -0.2 0 February 49 52.1 -0.6 -2.5 March 51 52.7 -0.7 -1 April 53 53 -1.4 1.4 May 55 53.4 -0.4 2 June 56 53.6 0.4 2.1 July 56 53.8 0.2 2 August 56 54.2 1.1 0.7 September 53 54.6 0.6 -2.2 October 54 54.8 -0.2 -0.6 November 55 54.9 0.6 -0.5 December 55 55.2 0.7 -0.8
January 54 55.5 -0.2 -1.3 February 55 56.1 -0.6 -0.5 March 55 56.9 -0.7 -1.1 April 54 57.6 -1.4 -2.1 May 57 58.2 -0.4 -0.8 June 60 58.8 0.4 0.8 July 61 59.4 0.2 1.4 August 65 60 1.1 3.9 September 62 60.6 0.6 0.8 October 62 61 -0.2 1.1 November 62 61.4 0.6 0.1 December 62 61.5 0.7 -0.2
January 62 61.5 -0.2 0.7 February 62 61.3 -0.6 1.3 . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 March 61 61.2 -0.7 0.5 April 59 61.2 -1.4 -0.7 May 60 61.2 -0.4 -0.8 June 60 61.3 0.4 -1.6 July 61 61.3 0.2 -0.5 August 61 61.3 1.1 -1.4 September 62 61.4 0.6 0 October 62 61.7 -0.2 0.5 November 62 62 0.6 -0.6 December 64 62.3 0.7 1
January 62 62.6 -0.2 -0.4 February 62 62.9 -0.6 -0.3 March 63 63.3 -0.7 0.4 April 63 63.6 -1.4 0.8 May 64 64 -0.4 0.4 June 64 64.2 0.4 -0.6 July 64 64.5 0.2 -0.7 August 66 65 1.1 -0.1 September 66 65.4 0.6 0 October 65 65.8 -0.2 -0.6 November 66
December 67
January 67 -0.2 February 68 -0.6 March 67 -0.7 April 67 -1.4
Analysis of the Time Series Decomposition Results
In this Section, we make a brief analysis of the behavior of each of the five time series and its constituent components. Results of more detailed investigation and analysis have been presented in Section 4.
Indian IT sector time series:
The Indian IT sector time series in Figure 1 depicts that from January 2009 till July 2011 the time series experienced a modest rate of growth. However, from August 2011 till November 2013 the time series had been rather sluggish with occasional decrease in its values. From December 2013 to October 2015 the IT sector time series index have consistently increased again, before experiencing stagnation again from December 2015 till April 2016. The behavior of the trend component of the IT sector time series can be J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 observed from Figure 6 and Table 1. The trend increased at a slow rate from January 2009 till February 2011. However, the trend started experiencing a fall from March 2011 and the downward trend continued till February 2012 before it started increasing again from March 2012. The trend increased at a very slow rate till February 2013 before it picked up a faster rate of increase from March 2013. However, the trend again started stagnating from March 2015 which continued till October 2015, which is the last trend figure that we could obtain in our study. We can also see the behavior of the seasonal component of the IT sector time series in Figure 6 and Table 1. It is observed that IT sector has a positive seasonal effect during September to March, while the seasonality impact is negative during April to July. The month of February has the highest positive seasonality in the time series, while the month of May has the most negative seasonality component. It is also observed that both the seasonal and the random components have very less magnitudes as compared to the trend component in the time series.
Indian CG sector time series:
The Indian CG sector time series experienced quite a large number of trend reversals as can be observed from Figure 7 and Table 2. However, roughly, we can divide the time series in four broad time horizons – (i) January 2009 – October 2010, a period during which the CG sector has experienced an upward movement, (ii) November 2010 – September 2013, when the sector has undergone a fall, (iii) October 2013 – July 2015, a period during which the sector had a rise again, and (iv) August 2015 – April 2016, when the sector witnessed a fall again which continued till the end of the time horizon under our study. The trend component of the time series also followed the same pattern. From Table 2, it may be easily seen that the CG sector has a positive seasonality effect during the months of June, July and August with the highest positive seasonality being found in the month of July. The seasonality is negative during the months of January to May, with the most negative value occurring in the month of February. The random component, in general, has more dominant presence than the seasonal component. However, as in the IT time series, the trend is the most predominant component in the CG time series.
DJIA index time series:
It is evident from Figure 3 and Table 3 that the DJIA time series index consistently increased during the entire period of our study, i.e., January 2009 – April 2016. A careful look at the trend component in Figure 7 makes it evident that the trend stagnated from January 2015 till October 2015 – the last month for which the trend values could be computed. From Table 3, it is also clear that DJIA index have positive seasonality during January to May. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 However, the months of June to November experience negative seasonal effects for DJIA index. The random component values are usually higher than those of the seasonal components. However, as in the India IT and Indian CG sector time series, the trend is the most dominant component in the DJIA index time series.
NIFTY index time series:
As it can be observed from Figure 4, the NIFTY time series has a number spikes and falls. However, we can broadly divide the time horizon into four divisions based on the behavior of the time series: (i) from January 2009 till October 2010, the NIFTY index had an overall upward movement, (ii) during November 2010 to August 2013 there was no substantial change in the NIFTY index values, (iii) from September 2013 to February 2015, the NIFTY index had again increased consistently, and (iv) during March 2015 to April 2016, the NIFTY index experienced a consistent fall. The trend component of the NIFTY time series exhibited the same behavior as can be observed from Figure 7. From Table 4, it is easy to observe that the seasonal component values are very nominal for the NIFTY time series. The month of October experiences the highest positive seasonality while the maximum negative seasonality is observed during the month of June. It is also clear from Table 4 that the random component values are more dominant than those of the seasonal component, while trend is the strongest component in the aggregate time series of NIFTY.
US Dollar to Indian Rupee exchange rate time series:
Figure 5 depicts the time series for the US Dollar to Indian Rupee exchange rates for the period January 2009 to April 2016. Again, based on the behavior of the time series, the time horizon can be divided into four intervals: (i) from January 2009 to August 2011, during which the exchange rate exhibited a slight downward trend, (ii) from September 2011 to May 2012, the period that experienced a moderate increase in the exchange rate, (iii) from June 2012 to May 2013, during which the exchange rate almost remained constant, and (iv) from June 2013 to April 2016, a period during which the exchange rate increased consistently. From Table 5 and Figure 10, it is evident that the trend of the time series also exhibited similar behavior. The seasonal and the random components in the time series are found to have negligible values compared to those of the trend signifying that the time series is predominantly composed of the trend component only. J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 Association Analysis of the Time Series
In order to investigate further into the behavior of the five time series, we carry out several experiments. In this Section, we discuss the details of the studies that we carried out and present the results obtained. The experiments that we have carried out can be broadly categorized into two groups: (i) association analysis of the Indian IT sector time series with the time series of DJIA, NIFT and the Dollar to Rupee exchange rate, (ii) association analysis of the Indian CG sector time series with the time series of DJIA, NIFT and the US Dollar to Indian Rupee exchange rate. This is driven by our two hypotheses: (i) The Indian IT sector is dependent on the overall world economy, and hence the IT time series is expected to be strongly coupled with the DJIA and the Dollar to Rupee exchange rate time series. IT time series is also expected to be strongly associated with the NIFTY time series since the stock prices of some of the blue chip IT companies (e.g., Tata Consultancy Services, Infosys Ltd etc.) have strong impacts on the NIFTY index values. (ii) The CG sector of India is based on India’s growth story and hence the CG sector time series is expected to have a strong association with the NIFTY index values. Since the DJIA index and the US Dollar to Indian Rupee exchange rate are related to the world economy, the CG sector time series of India is expected to have a very less association with these two time series. In order to verify the above two hypotheses, we carry out bivariate correlation tests and cross correlation tests (Shumway & Stoffer, 2011) of the both the IT time series and the CG time series with the DJIA, NIFTY and the US Dollar to Indian Rupee exchange rate time series. In Section 4.1 and Section 4.2, we present the detailed results of the studies of the Indian IT sector and the Indian CG sector with respect to their associations with the DJIA, NIFTY and the Dollar to Rupee exchange rates.
Association Analysis of the Indian IT Time Series
We have tested association of the IT time series with each of the time series of DJIA, NIFTY and Dollar to Rupees exchange rate time series. The detailed results of the study are discussed in this Section. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 Association between the Indian IT and DJIA time series
In order to study the association between the Indian IT sector time series and the DJIA index time series, we have first plotted the aggregate time series of both the sectors for the period January 2009 to April 2016. Figure 11 depicts the plot. It can be easily observed that the two time series exhibited similar behavior during the period under study.
Figure 11
Comparison of the aggregate time series of the DJIA index and the Indian IT sector index (Jan 2009 – Apr 2016) We have also studied the behavior of the trend components of the two time series during the same period. Figure 12 presents the trend plots of the two time series. It is evident that the trends of the IT time series and the DJIA time series behaved in an identical manner during January 2009 to April 2016. In the similar way, we made a comparative analysis of the behavior of the seasonal components of the two time series. Figure 13 depicts the results obtained. It is clear that the seasonal components of the two time series exhibited similar behavior with the Indian IT seasonality having a lag with respect to the DJIA seasonality. The results in Figures 11, 12 and 13 clearly indicate that Indian IT sector time series has a strong association with the DJIA index time series. J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Figure 12
Comparison of the trend components of the DJIA index and the Indian IT sector index time series (Jan 2009 – Apr 2016)
Figure 13
Comparison of the seasonal components of the DJIA index and the Indian IT sector index time series (Jan 2009 – Apr 2016) Having established graphically that the Indian IT sector time series has a strong association with the DJIA time series, we carry out some statistical tests in R in order to prove the association using formal computations. We use cor.test ( ) function in R to carry out a bivariate correlation test between the IT time series and the DJIA time series. The results of the test are presented in Table 6. The high value (0.945425) of the correlation coefficient with a negligible p-value of the Null hypothesis (of no correlation) implies that the two time series are highly correlated on point-to-point basis. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 Table 6
Results of the correlation test between Indian IT sector and the DJIA index
Parameter Value t- statistic Degrees of freedom (df) Significance values (p - value) Correlation coefficient 26.907 86 < 2.2e-16 0.945425
We also study the values of the correlation coefficient at different lags of IT time series with respect to the DJIA time series in order to identify which lag yields the highest value of the correlation coefficient. We used the ccf ( ) function in R programming environment for this purpose. Figure 14 presents the results. It is evident from Figure 14 that at lag = 0 the highest value of the correlation coefficient is achieved. Hence, it is concluded that the Indian IT time series and the DJIA time series are highly correlated with a zero lag in between them.
Figure 14
The output of the ccf( ) function depicting the cross correlation between aggregate IT time series and the aggregate DJIA time series (Jan 2009 – Apr 2016)
Table 7
Correlation test for the seasonal components of the Indian IT index and the DJIA index
Parameter Value t- statistic Degrees of freedom (df) Significance values (p - value) Correlation coefficient -0.47595 86 0.6353 -0.05125503
We also studied the association between the seasonal components of the Indian IT time series and the DJIA time series. The results obtained in bivariate correlation test using the cor.test ( ) function in R environment are presented in Table 7. The extremely small value of the correlation coefficient and the high value of significance indicate that there is no point to point correlation between the seasonal components of the two time series. J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
In order to identify the lag at which seasonal components of the two time series attain the highest value of the correlation coefficient, we use the ccf( ) function in R. Figure 15 presents the results. It may be observed that although the correlation is very low (-0.05) at lag value of zero, the cross correlation is approximately around 0.9 (which is quite high) at a lag value of 0.25. Since a lag of 1 represents a time horizon of 12 months, a lag of 0.25 is equivalent to 3 months duration. In other words, the seasonal component of the Indian IT time series has a very strong correlation (e.g., correlation coefficient value approximately 0.9) with the seasonal component of the DJIA time series with a lag of 3 months.
Figure 15
The output of the ccf () function depicting the cross correlation between the seasonal components of the IT time series and the DJIA time series
Association between Indian IT sector index and Dollar to Rupee exchange rate
In order to study the association between the IT sector time series and the US Dollar to Indian Rupee exchange rate time series, we have first plotted the aggregate time series of both the sectors for the period January 2009 to April 2016. Figure 16 depicts the plot. We have also studied the behavior of the trend components of the two time series of Indian IT sector and the Dollar to Rupee exchange rate. Figure 17 presents the comparison of the trend components. It may be noted that both in Figure 16 and 17, we have multiplied the Dollar to Rupee exchange rate by a factor 0f 100 before plotting in order to make a parity between the ranges of values of the two time series for the purpose of comparison. We do not carry out any study on the seasonality components since for the exchange rate time series, the seasonality does not make any sense. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 Figure 16
Comparison of the aggregate time series of the Indian IT sector index and the US Dollar to Indian Rupees exchange rate (Jan 2009 – Apr 2016)
Figure 17
Comparison of the trend components of the Indian IT sector index and the US Dollar to Indian Rupees exchange rate time series (Jan 2009 – Apr 2016) As per our hypotheses, we expect a strong association between the IT time series and the Dollar to Rupee exchange rate time series. A cursory visual inspection of the graphs in Figure 14 and Figure 15 enables us to see a positive association between the two time series. However, we carried out correlation and cross-correlation tests to compute the quantitative values of the association. The results obtained in bivariate correlation test using the cor.test ( ) function in R environment are presented in Table 8. The high value (0.8333524) of the correlation coefficient with a negligible p-value of the null hypothesis (of no correlation) implies that the two time series are highly correlated on point-to-point basis. J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Table 8
Correlation test for the Indian IT index and the US Dollar to Indian Rupee exchange rates
Parameter Value t- statistic Degrees of freedom (df) Significance values (p - value) Correlation coefficient 13.982 86 < 2.2e-16 0.8333524
We also carried out correlation test between the trend components of the IT sector time series and the Dollar to Rupee exchange rate time series. The test yielded a high value (0.8794656) of correlation coefficient with a negligible value of significance level (i.e., the p-value) of less than 2.2 e-15. This clearly indicated a strong point-to-point positive correlation between the two time series and thereby validated our hypothesis that Indian IT sector time series is strongly coupled with the Dollar to Rupee exchange rate time series, both reflecting the world economic picture.
Figure 18
The output of the ccf ( ) function depicting the cross correlation between aggregate IT time series and the aggregate Dollar to Rupee exchange rate In order to identify the value of the lag that attains the largest magnitude of the correlation coefficient between the IT time series and the Dollar to Rupee exchange rate time series, we used the ccf ( ) function in R. Figure 16 presents the results. It is evident from Figure 16 that at lag = 0 the highest value of the correlation coefficient is achieved. We also computed the cross-correlation between the trend components of the two time series and also observed that the highest value of the correlation between the time series was obtained at a lag = 0. Hence, it is concluded that the Indian IT time series and the Dollar to Rupee exchange rate time series are highly correlated with a zero lag in between them. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 Association between Indian IT and NIFTY index time series
The plots of the aggregate time series, the trend components and the seasonal components of the Indian IT sector and the NIFTY index are depicted in Figure 17, Figure 18 and Figure 19 respectively. Even a visual inspection of Figure 17 and Figure 18 gives us an idea that there is a positive association between the IT sector index time series and the NIFTY time series and between their trend components. This validates our hypothesis that the Indian IT sector blue chip stocks have a strong impact on the NIFTY index values, which leads to a positive association between the two index. However, as in the previous cases, we validate our hypothesis by carrying out bivariate correlation tests and cross-correlation tests. Table 9 presents the results of correlation test on the IT sector time series and the NIFTY index time series using the cor.test ( ) function in R. The high value ( ) of the correlation coefficient with a negligible p-value of the null hypothesis (null hypothesis assumes no correlation) implies that the two time series are highly correlated on point-to-point basis. A correlation test is also carried out between the trend components of the IT sector time series and the NIFTY index time series. The test yielded a even higher value (0.9849783) of correlation coefficient with a negligible value of significance level (i.e., the p-value) of less than 2.2 e-16. This clearly indicated a strong point-to-point positive correlation between the two time series and thereby validated our hypothesis that Indian IT sector time series is strongly coupled with the NIFTY index time series. The cross-correlation study was carried out using the ccf ( ) function in R. Figure 17 presents the results of the cross-correlation between the IT sector time series and the NIFTY index time series. It is evident that at lag value of zero the highest correlation is attained. This makes it evidently clear that the Indian IT sector and the NIFY index have a very strong point-point positive association between them.
Figure 19
Comparison of the aggregate time series of the Indian IT sector index and the NIFTY index (Jan 2009 – Apr 2016) J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Figure 20
Comparison of the trend components of the Indian IT sector and the NIFTY index time series (Jan 2009 – Apr 2016)
Table 9
Correlation test for the Indian IT index and the NIFTY index
Parameter Value t- statistic Degrees of freedom (df) Significance values (p - value) Correlation coefficient 32.202 86 < 2.2e-16 0.9609465
Figure 21
Comparison of the seasonal components of the Indian IT sector and the NIFTY index time series (Jan 2009 – Apr 2016) . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 In order to investigate the association between the seasonal components of the IT sector time series and the NIFTY index time series, we carry out a correlation and cross correlation study on the seasonal component values of the two time series. Table 10 and Figure 20 present the results of the correlation test and the cross-correlation test.
Table 10
Correlation test for the seasonal components of the Indian IT index and the NIFTY index
Parameter Value t- statistic Degrees of freedom (df) Significance values (p - value) Correlation coefficient 2.5203 86 0.01357 0.2622572
The very low value of the correlation coeffieicent in Table 10 and low p-value indicate that the sesonal components of the IT sector and the NIFTY index have poor point-to point correlation. Figure 20 depcits the output of the ccf ( ) function in R and presents the cross-correlation results. It is evident that at lag = 0 the sesonal components have a very low correlation of 0.26. The highest correlation value of approximately 0.47 is achieved at the lag of - 0.4, which is equivalent to 5 months. Both the correlation test and the cross-correlation test results indicate a poor association between the sesonality of the IT sector and the NIFTY index.
Figure 22
The output of the ccf ( ) function depicting the cross correlation between the seasonal components of the IT sector and the NIFTY time series index J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
We have tested the association of the Indian CG sector with the DJIA, NIFTY and Dollar to Rupees exchange rate time series in the same way as we have done it for the Indian IT sector in Section 4.1. The detailed results of the study are discussed in this Section.
Association between the Indian CG and the DJIA time series
In Figure 21, Figure 22 and Figure 23, we have plotted the Indian CG sector and the DJIA aggregate time series, their trend components and their seasonal components respectively.
Figure 23
Comparison of the aggregate time series of the Indian CG sector index and the DJIA index (Jan 2009 – 2016) It is quite evident by visual inspection of Figure 21 and Figure 22 that there is no clear association between the aggregate time series of the two sectors as well as in the time series of their trend components. As in all analyses for the IT sector time series, we carry out the bivariate correlation test and the cross-correlation test for the aggregate time series of the CG sector and the DJIA index. Table 11 and Figure 24 present respectively present the results of the correlation test and the cross-correlation test. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Figure 24
Comparison of the trend components of the Indian CG sector index and the DJIA index (Jan 2009 – Apr 2016)
Figure 25
Comparison of the seasonal components of the Indian CG sector index and the DJIA index (Jan 2009 – Apr 2016)
Table 11
Correlation test for the Indian CG sector index and the DJIA index
Parameter Value t- statistic Degrees of freedom (df) Significance values (p - value) Correlation coefficient 0.69046 70 0.4922 0.08224673 J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Figure 26
The output of the ccf ( ) function depicting the cross correlation between aggregate CG time series and the aggregate DJIA index time series The extremely low value of correlation coefficient (0.08224673) in Table 11 clearly indicates that there is effectively no association between the aggregate time series of the CG sector and the DJIA index. From Figure 24, it is also evident that the numerically largest value of the correlation between the CG and the DJIA aggregate time series is attained at a lag value of nine months. The value of the maximum correlation being -0.32, at a lag of nine months again indicated an absence of any association between the time series. We also carried out the correlation and the cross-correlation tests between the trend components of the two time series. The correlation coefficient between the trend components has been found to be 0.001 with a p-value of 0.05533. The cross-correlation study found that the highest value of correlation was at a lag of 15 months and the value of the correlation coefficient at that lag was approximately 0.27. All these observations indicate that the association between the CG sector time series and the DJIA time series is very poor. This validates our hypothesis that CG sector India is tied to India’s domestic growth story and it should have a minimal association with the pattern of growth in DJIA index which essentially reflects the world growth story. Since the aggregate time series and the trend components of the Indian CG sector index and the DJIA index were found to have a very poor association, we did not carry out any study on their seasonal components. However, even a very casual look at Figure 23 will make it clear that the seasonal components of the two time series have no association among them. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Association between Indian CG sector index and Dollar to Rupee exchange rate
As in all the cases that we have discussed so far, we first plot the behavior of the aggregate time series and their trend components for the Indian CG sector and the US Dollar to Indian Rupee exchange rates. Note that seasonality does not make any sense in the Dollar to Rupee exchange rate time series and hence we do not carry out any seasonality analysis for this comparative study of two time series. Figure 25 and Figure 26 depict respectively the aggregate time series and their trend components for the Indian CG sector index and the Dollar to Rupee exchange rates.
Figure 27
Comparison of the aggregate time series of the Indian CG sector index and US Dollar to Indian Rupee exchange rate (Jan 2009 – Apr 2016) As in the case of comparison of Indian IT sector index time series with the US Dollar to Indian Rupee exchange rate, in both Figure 25 and Figure 26, we have multiplied the Dollar to Rupee exchange rate by a factor of 100 before plotting in order to make a parity between the ranges of values of the two time series for the purpose of comparison. We do not carry out any study on the seasonality components since for the exchange rate time series, the seasonality does not make any sense. Even though a visual inspection of Figure 25 and Figure 26 clearly elicit the fact that there is no perceptible association between the Indian CG sector index time series and the US Dollar to Indian Rupee time series and also between their trends, we carry out bivariate correlation and cross-correlation studies to in order to find the strength of their association. Table 12 and Figure 27 present the results of the correlation and cross-correlation studies respectively. J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Figure 28
Comparison of the trend components of the Indian CG sector index and the US Dollar to Indian Rupee exchange rate (Jan 2009 – Apr 2016)
Table 12
Correlation test for the Indian CG index and the US Dollar to Indian Rupee exchange rate
Parameter Value t- statistic Degrees of freedom (df) Significance values (p - value) Correlation coefficient 1.028 86 0.3068 0.1101802
Figure 29
The output of the ccf ( ) function depicting the cross correlation between aggregate CG time series and the aggregate Dollar to Rupee exchange rate time series From Table 12, we observe that the value of the correlation coefficient (0.11018) is too low with a p-value of 0.3068. Since the p-value is much larger than the threshold value of 0.05, the null hypothesis that assumed no correlation between the time series has got a good support. Hence, the low value of the correlation coefficient and an associated high p-value indicate that . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 the association between the CG sector time series and the US Dollar to Indian Rupee exchange rate is very poor. Figure 27 shows that the highest correlation is attained at a lag of 5 months with the maximum value of the correlation being 0.15. We also carried out correlation and cross-correlation tests on the trend components of the two time series. It has been observed that the trend components of the CG sector and the exchange rate values have a correlation of 0.0951472 with a p-value of 0.4136.The cross-correlation study indicated that the highest correlation was found at a lag of 15 months with the maximum value of the correlation being -0.45. All these observations strongly support our hypothesis that there is no association between the Indian CG sector index time series and the US Dollar to Indian Rupee exchange rate – the CG sector depicting the Indian domestic growth story, while the Dollar to Rupee exchange rate presenting the world economic story.
Association between Indian CG sector and NIFTY index time series
We investigate the degree of association between the Indian CG sector index time series and the NIFTY index time series. Our hypothesis is that both the CG sector and the NIFTY relate to the Indian growth story and it is natural that there would a high degree of association between them. We carried out graphical analysis of the two time series by plotting their aggregate index, the trend components and the seasonal components in Figure 28, Figure 29 and Figure 30 respectively.
Figure 30
Comparison of the aggregate time series of the Indian CG sector index and the NIFTY index (Jan 2009 – Apr 2016) Although, from Figure 28 and Figure 29, it possible to get an idea that there is an association between the aggregate time series of the CG sector and the NIFTY index as well as between their trend components, we carried out bivariate correlation and cross-correlation J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 analysis between these time series in order to compute the degree of association between them. Table 13 and Figure 31 present the results of the correlation test and the cross-correlation test respectively.
Table 13
Correlation test for the Indian CG sector and the NIFTY index time series
Parameter Value t- statistic Degrees of freedom (df) Significance values (p - value) Correlation coefficient 8.8156 86 1.168e-13 0.6889807
Figure 31
Comparison of the trend components of the Indian CG sector and the NIFTY index time series (Jan 2009 – Apr 2016) From Table 12, it may observed that the value of the correlation coefficient between the overall time series of the CG sector and the overall time series of the NIFTY index is 0.688987 with an associated p-value of which is negligible. Since the p-value indicates the level of support for the null hypothesis that assumes no correlation among the two time series, a reasonably high value of the correlation coefficient and a negligible p-value both indicate a strong association between the two time series. From Figure 31, we observe that the maximum value of correlation coefficient (0.688987) between CG time series and the NIFTY time series is attained at a lag value of zero. This supports the fact that the CG sector time series and the NIFTY index time series has a high point-to-point correlation and hence a strong association between them. We also carried out correlation and cross-correlation tests on the trend components of the two time series. The correlation tests between the trend components yielded a value of correlation coefficient of 0.6052231 with an associated p-value of 6.988e-09. The . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 reasonably high value of the correlation coefficient along with a negligible p-value again indicated a strong association between the trend components of the two time series. The cross-correlation analysis on the trend components showed that the highest value of the correlation (0.6052231) between the trend components of the two time series was at a lag value of zero. All these observations clearly indicated that the CG sector time series and the NIFTY time series have a very strong association between them. This supports our hypothesis that both these sectors essentially depict the domestic growth story of India and hence they should inherently have a strong degree of association among them.
Figure 32
Comparison of the seasonal components of the Indian CG sector and the NIFTY index time series (Jan 2009 – Apr 2016) Since the CG sector time series and the NIFTY index time series and the time series of their trend components are found to have a strong association, we also studied the association between their seasonal components. The visual analysis of Figure 30 being a bit difficult proposition, we carried out correlation and cross-correlation tests on the seasonal components of the two time series. The correlation test yielded a value of 0.2353171 for the correlation coefficient with an associated p-value of 0.02731.These values indicated that the seasonal components of the two time series have a poor point to point correlation. However, as depicted in Figure 32, the cross-correlation study revealed the fact that the CG sector seasonal components has a maximum value of correlation coefficient with the seasonal components of the NIFTY sector at a lag of 4 months with the maximum correlation value being approximately 0.7. This implies that the seasonal components of the two time series have high correlations at certain lags albeit a low point-to-point correlation between them. This further supports our J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 hypothesis of strong association between the Indian CG sector index time series and the NIFTY index time series.
Figure 33
The output of the ccf ( ) function depicting the cross correlation between aggregate CG time series and the aggregate NIFTY index time series
Figure 34
The output of the ccf ( ) function depicting the cross correlation between the seasonal components of the Indian CG sector and the NIFTY time series index . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 Forecasting of Time Series Using Linear Regression Model
In Section 4, we have analyzed association among various sectors based on their time series index values. We have found that Indian IT sector index time series has a strong association with the DJIA index, the US Dollar to Indian Rupee exchange rate time series, and also with the NIFTY index time series. While analyzing the Indian CG sector behavior, we observed that the Indian CG sector has a strong association with NIFTY index, but does not have any association with the DJIA index and the US Dollar to Indian Rupee exchange rate. In this Section, we take our analysis to the next level by formulating forecasting models based on linear regression analysis for the time series which exhibited strong association among themselves. We build the following linear regression models for forecasting: 1.
Regression of the Indian IT sector index time series on the DJIA index time series and vice-versa. 2.
Regression of the Indian IT sector index time series on the US Dollar to Indian Rupee exchange rate time series and vice versa. 3.
Regression of the Indian IT sector time series on the NIFTY time series and vice versa. 4.
Regression of the Indian CG sector time series on the NIFTY time series and vice versa. Note that we do not build any model for the Indian CG sector with DJIA index and the CG sector with Dollar to Rupee exchange rate since these pairs of time series were found to have no associations between in them as per our analysis in Section 4.2.1 and Section 4.2.2. For the purpose of building the linear models between any pair of time series, we suitably divide our datasets into two disjoint sets – one serving as the training data set to construct the model and the other is the test data set to test the effectiveness of the model that is built. For the purpose of training data set construction for any time series, we use the time series data from January 2009 to December 2014. The time series data for the reaming period, i.e., from January 2015 to April 2016 are used for the purpose of test data set. Based on the linear model that is constructed from the training data sets, the values of the dependent time series are forecasted and then compared with the actual values in the test data set. The forecasting error is computed for each month so as to have an idea about the effectiveness of the regression model. In the following sub-sections, we clearly describe the methodology that we have followed in building and testing the forecasting models. J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Regression model of Indian IT sector index with the DJIA index time series
In Section 4.1.1, we observed that the bivariate correlation between the Indian IT sector index time series and the DJIA index time series was 0.945425. This indicated that there is a strong linear relationship between the two time series. We now attempt to construct two linear regression models – one for the regression of the Indian IT sector index onto the DJIA index and the other for the regression of the DJIA index onto the Indian IT sector index. While the first model will use the DJIA index as the independent variable and hence will be used for forecasting the Indian IT sector index, the second model will be built using the Indian IT sector index as the independent variable and would be used for forecasting the DJIA index. We use the time series data for both the DJIA index and the Indian IT sector index from January 2009 to December 2014 as the training data set for constructing the model. The time series data from January 2015 to April 2016 are used as the testing data set for evaluating the model efficiency by computing the forecasting error. For constructing the linear model of the Indian IT sector index on the DJIA index we use the lm (it_timeseries ~ djia_timeseries) function call, where the it_timeseries parameter stands for the R variable that stores the Indian It sector time series index from January 2009 to December 2014, and the djia_timeseries parameter stands for the time series variable that stores the DJIA index for the same period of January 2009 to December 2014. The lm ( ) function call constructs a linear regression model with the parameter at the left hand side of the formula operator ~ being used as the dependent variable, and the parameter at the right hand side of the operator ~ as the independent variable. Using the summary ( ) function in R, we found some important parameters of the linear regression model of Indian IT index onto the DJIA index. The results are presented in Table 13.
Table 14
Summary results of the linear model of the Indian IT index onto the DJIA index
Parameter Value
Constant term in the regression model -2585.21 Coefficient of the DJIA index 0.6958 Residual standard error 752.4 on 70 degrees of freedom Multiple R- squared 0.8686 Adjusted R-squared 0.8667 . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Since the adjusted R-squared value is quite high, we use the linear model to forecast the Indian IT sector index for the period January 2015 to April 2016. We do not bring in any other lag values in the cross-correlation function results since the lag = 0 has a very high point-to-point correlation coefficient between the two time series. We also compare the forecasted IT index with their actual values for each month and compute the forecast error. The results are presented in Table 14. It may be noted that the percentage values of errors are quite nominal.
Table 15
Results of forecasting of Indian IT sector index based on DJIA index
Year Month Actual IT Index (A) Actual DJIA Index (B) DJIA Index * Coefficient (C) Constant Term (Intercept) (D) Forecasted IT Index E = (C + D) Percent Error (|E – A|/A) *100
In a similar manner, we constructed a linear model of DJIA index onto the Indian IT sector index for the purpose of forecasting DJIA index based on the Indian IT sector index values. Table 15 presents the output of the summary ( ) function in R for this model. Since the adjusted R-squared value is very high, we use this linear model for forecasting the DJIA index for the period January 2015 to April 2016, based on the Indian IT sector index values. The results are presented in Table 16. It may be noted that the magnitudes of the error percentages are quite acceptable. J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Table 16
Summary results of the linear model of the DJIA index onto the Indian IT index
Parameter Value
Constant term in the regression model 4899.25 Coefficient of the IT sector index 1.248 Residual standard error 1008 on 70 degrees of freedom Multiple R- squared 0.8686 Adjusted R-squared 0.8667
Table 17
Results of forecasting of the DJIA index based on the Indian IT sector index
Year Month Actual DJIA Index (A) Actual IT Index (B) Actual IT Index * Coefficient (C) Constant Term (Intercept) (D) Forecasted DJIA Index E = (C + D) Percent Error (|E – A|/A) *100
Regression model of the Indian IT sector index with Dollar to Rupee exchange rate
In Section 4.1.2, we observed that the Indian IT sector index time series had a strong positive association with the US Dollar to Indian Rupee exchange rate time series. The correlation coefficient between the two time series was found to be
Hence, we build two linear models for these two time series – one for the regression of Indian IT sector index onto the Dollar to Rupee exchange rate, and the other for the regression of Dollar to Rupee exchange rate onto the Indian IT sector index. Since the methodology for building the models remains identical to the one discussed in Section 5.1, we present only the results of the two models. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Table 17 presents the output of the summary ( ) function for the linear model of the Indian IT sector index time series onto the Dollar to Rupee exchange rate time series.
Table 18
Summary results of the linear model of the Indian IT sector index onto the US Dollar to Indian Rupee exchange rate
Parameter Value
Constant term in the regression model -5722.8 Coefficient of the Dollar to Rupee exchange rate 228.3 Residual standard error 1451 on 70 degrees of freedom Multiple R- squared 0.5113 Adjusted R-squared 0.5045
Table 19
Results of forecasting of Indian IT sector index based on US Dollar to Indian Rupee exchange rates
Year Month Actual IT Index (A) Actual Exchange Rate (B) Actual Exch Rate * Coefficient (C) Constant Term (Intercept) (D) Forecasted IT Index E = (C + D) Percent Error (|E – A|/A) *100
The results of forecasting for the Indian IT sector index for the period January 2015 to April 2016 based on the US Dollar to Indian Rupee exchange rate have been presented in Table 18. The errors in forecasting are found to be quite nominal considering the fact that exchange rate time series has a small range of variations compared to the variations in the Indian IT sector index value.
We also constructed a linear model of the US Dollar to Indian Rupee exchange rate time series on to the Indian IT sector index time series.
Table 19 presents the output of the summary ( ) function for the linear model. This model has been used to forecast the Dollar to Rupee exchange rates based on the Indian IT sector index values. J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Table 20
Summary results of the linear model of the US Dollar to Indian Rupee exchange rate onto the Indian IT sector index
Parameter Value
Constant term in the regression model 38.50 Coefficient of the IT sector index 0.002240 Residual standard error 4.546 on 70 degrees of freedom Multiple R- squared 0.5113 Adjusted R-squared 0.5043
Table 21
Results of forecasting of the US Dollar to Indian Rupee exchange rate based on Indian IT sector index
Year Month Actual Exchange Rate (A) Actual IT Index (B) Actual IT Index * Coefficient (C) Constant Term (Intercept) (D) Forecasted Exch. Rate E = (C + D) Percent Error (|E – A|/A) *100
Since the adjusted R-squared value is 0.5043, we investigated whether there was any possibility to increase it by bringing in values at some other lags. However, we found that by bringing in time series values at other lags, we could not improve the adjusted R-squared value. Hence we used the model in Table 19 for forecasting the US Dollar to Indian Rupee exchange rates for the period January 2015 to April 2016 based on the Indian IT sector index during the same period. The results are presented in Table 20. The percentage values of the forecast errors are found to be quite low. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Regression model of Indian IT sector index with NIFTY index
In Section 4.1.3, we observed that the Indian IT sector index had a very strong positive association with the NIFTY index. The correlation coefficient between the two time series was found to be 0.9609465. Therefore, we build to linear models for these two time series. Table 21 presents the output of the summary ( ) function of the linear model of the Indian IT sector index onto the NIFTY index. We will use this model to forecast IT sector index for the period January 2015 to April 2016 based on the NIFTY index values during the same period.
Table 22
Summary results of the linear model of the Indian IT index onto the NIFTY index
Parameter Value
Constant term in the regression model -3179.63 Coefficient of the NIFTY index 1.685 Residual standard error 709.6 on 70 degrees of freedom Multiple R- squared 0.8831 Adjusted R-squared 0.8815
Table 23
Results of forecasting of the Indian IT sector index based on the NIFTY index
Year Month Actual IT Index (A) NIFTY Index (B) NIFTY Index * Coefficient (C) Constant Term (Intercept ) (D) Forecasted IT Index E = (C + D) Percent Error (|E – A||/A) *100 J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Since the adjusted R-squared value in Table 21 is quite high, we used the linear model in Table 21 without investigating any other lags in the cross-correlation function. The linear model is used for forecasting Indian IT sector index for the period January 2015 to April 2016 based on the NIFTY index during that period. The forecasting results are presented in Table 22. It is evidently clear that the percentage values of the forecast errors are quite small. We also built a linear model of the NIFTY index onto the Indian IT sector index time series. Table 23 depicts the output of the summary ( ) function of the model. We have used this model to forecast the NIFTY index for the period January 2015 to April 2016 based on the Indian IT sector index during the same period.
Table 24
Summary results of the linear model of the NIFTY index onto the Indian IT index
Parameter Value
Constant term in the regression model 2321.47 Coefficient of the IT sector index 0.524 Residual standard error 395.7 on 70 degrees of freedom Multiple R- squared 0.8831 Adjusted R-squared 0.8815
Table 25
Results of forecasting of the NIFTY index based on the Indian IT sector index
Year Month Actual NIFTY Index (A) Actual IT Index (B) Actual IT Index * Coefficient (C) Constant Term (Intercept) (D) Forecasted NIFTY Index E = (C + D) Percent Error (|E – A|/A) *100 . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Since the linear model has yielded an adjusted R-squared value that is quite high, we do not consider any other lag in the cross-correlation function between the two time series and use the linear model in Table 23 to forecast the NIFTY index based on the Indian IT sector index. The results of forecasting are presented in Table 24. The percentage values of the forecast errors are found be quite low and hence acceptable.
Regression model of the Indian CG sector index with the NIFTY index
In Section 4.2.3, we observed that the Indian CG sector index have a fair degree of association with the NIFTY index. The correlation coefficient between the two time series over the period January 2009 to April 2016 was found to be 0.6889807. Based on this fair degree of association between the two time series, we have built to linear models for the purpose of forecasting the values of one time series given the other and vice versa.
Table 26
Summary results of the linear model of the Indian CG index onto the NIFTY index (The training data set for the model is based on data for the period Jan 2009 – Dec 2014)
Parameter Value
Constant term in the regression model 4756.20 Coefficient of the NIFTY index 1.2444 Residual standard error 2251 on 70 degrees of freedom Multiple R- squared 0.2905 Adjusted R-squared 0.2804
Table 27
Results of forecasting of Indian CG sector index based on the NIFTY index (The model used data for the period Jan 2009 – Dec 2014 as the training data set)
Year Month Actual CG Index (A) Actual NIFTY Index (B) NIFTY Index * Coefficient (C) Constant Term (Intercept) (D) Forecasted CG Index E = (C + D) Percent Error (|E – A|/A) *100 J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Table 25 depicts the output of the summary ( ) function of the linear model of the Indian CG sector index onto the NIFTY index. We have used the time series data of both the time series for the period January 2009 to December 2014 for constructing the training data set of the model. Since the value of the adjusted R-squared in Table 25 is quite low, we investigated whether it could be improved upon by considering the time series values at other lags in the cross-correlation function. However, we failed in our attempt as the value of the adjusted R-squared could not be increased. Hence, we used the linear model in Table 25 to forecast the Indian CG sector index for the period January 2015 to April 2016 based on the NIFTY index over the same period. Table 26 presents the forecasting results. It is quite clear that the percentage values of forecast errors are low and within the acceptable range.
With the objective of forecasting the NIFTY index based on the Indian CG sector time series index, we build a linear model of the NIFTY index onto the Indian CG sector index. The output of the summary ( ) of this linear model is depicted in Table 27. Note that the training data set for the model is based on both the time series data for the period January 2009 to December 2014. We used the linear model in Table 27 to forecast the NIFTY index for the period January 2015 to April 2016 based on the Indian CG sector index over the same period.
Table 28
Summary results of the linear model of the NIFTY index onto the Indian CG index (The training data set for the model is based on data for the period Jan 2009 – Dec 2014)
Parameter Value
Constant term in the regression model 2867.72 Coefficient of the Indian CG sector index 0.2335 Residual standard error 974.9 on 70 degrees of freedom Multiple R- squared 0.2905 Adjusted R-squared 0.2804
Table 28 represents the forecasting results of the NIFTY index based on the India CG sector index for the period January 2015 to April 2016. The percentage values of the forecasting errors are found to be a bit high although within the acceptable range of less than 25%. Although the percentage values of forecasting errors in both Table 26 and Table 28 are within acceptable limits, we explored ways to improve upon the forecasting accuracies by suitably changing the model construction methods. A careful look at Figure 2 indicates that the CG index time series has undergone a number of changes in its pattern over the period January 2009 to April 2016. Similarly, the NIFTY index time series also has undergone a number of . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 changes in its behavior over the same period of time as evident from Figure 4. Linear models constructed from these changing time series over a long time horizon of 6 years (January 2009 to December 2014) will inherently be far from perfect. We observe that both the Indian CG sector index and the NIFTY index time series had experienced consistent upward trends during the entire year of 2014 and both have exhibited somewhat downward swings in the year of 2015 onwards. We utilized this observation in choosing our training data set for constructing the linear models between the CG sector index and the NIFTY index. For constructing the linear models, we leave out the time series index of both the sectors from January 2009 till December 2013 and use the data of 2014 as the training data set. The objective is to build the models based on the most recent observations leaving out past data over a long time horizon.
Table 29
Results of forecasting of the NIFTY index based on Indian CG sector index (The model used data for the period Jan 2009 – Dec 2014 as the training data set)
Year Month Actual NIFTY Index (A) Actual CG Index (B) Actual CG Index * Coefficient (C) Constant Term (Intercept) (D) Forecasted NIFTY Index E = (C + D) Percent Error (|E – A|/A) *100
Table 29 presents the output of the summary ( ) function of the newly constructed linear model of the Indian CG sector onto the NIFTY index. Note that the model is constructed using the time series index of the Indian CG sector and the NIFTY from January 2014 to December 2014 as the training data set. It is interesting to observe that the adjusted R-squared value has increased to 0.8322 from 0.2804 (from Table 25) just by changing the training data set to include only the most recent observations. The linear model in Table 29 is used for forecasting the Indian CG sector index for the period January 2015 to April 2016 based on the NIFTY J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 index for the same period. Table 30 presents the results of forecasting. A visual inspection of the error values in Table 26 and Table 30 makes it evidently clear that the forecast accuracy has improved substantially in the new model.
Table 30
Summary results of the linear model of the Indian CG index onto the NIFTY index (The training data set for the model is based on data for the period Jan 2014 – Dec 2014)
Parameter Value
Constant term in the regression model -7174.22 Coefficient of the NIFTY index 2.7870 Residual standard error 979.8 on 10 degrees of freedom Multiple R- squared 0.8474 Adjusted R-squared 0.8322
Table 31
Results of forecasting of the Indian CG sector index based on the NIFTY index (The model used data for the period Jan 2014 – Dec 2014 as the training data set)
Year Month Actual CG Index (A) Actual NIFTY Index (B) NIFTY Index * Coefficient (C) Constant Term (Intercept) (D) Forecasted CG Index E = (C + D) Percent Error (|E – A|/A) *100
Table 31 presents the output of the summary ( ) function for the new linear model of the NIFTY index onto the Indian CG sector index. The model is constructed using the time series index of the NIFTY index and the Indian CG sector index from January 2014 to December 2014 as the training data set. We observe that the adjusted R-squared value has increased to 0.8322 from . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Table 32
Summary results of the linear model of the NIFTY index onto the Indian CG index (The training data set for the model is based on data for the period Jan 2014 – Dec 2014)
Parameter Value
Constant term in the regression model 3323.79 Coefficient of the Indian CG sector index 0.3041 Residual standard error 323.6 on 10 degrees of freedom Multiple R- squared 0.8474 Adjusted R-squared 0.8322
The linear model in Table 31 is used for forecasting the NIFTY index for the period January 2015 to April 2016 based on the Indian CG sector index for the same period. Table 32 presents the results of forecasting. A visual inspection of the error values in Table 28 and Table 32 makes it evidently clear that the forecast accuracy has improved substantially in the new model.
Table 33
Results of forecasting of the NIFTY index based on Indian CG sector index (The model used data for the period Jan 2014 – Dec 2014 as the training data set)
Year Month Actual NIFTY Index (A) Actual CG Index (B) Actual CG Index * Coefficient (C) Constant Term (Intercept) (D) Forecasted NIFTY Index E = (C + D) Percent Error (|E – A|/A) *100 J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 Related Work
Researchers have spent considerable effort in designing mechanisms for forecasting of daily stock prices. Applications of neural network based approaches have been proposed in many forecasting systems. The main advantage of neural networks is that they can approximate any nonlinear function to an arbitrary degree of accuracy with a suitable number of hidden units (Hornik et al., 1989). Zhang et al. propose the application of a well-known neural network technique - multilayer back propagation (BP) neural network - in financial data mining (Zhang et al., 2004). The authors present a modified neural network-based forecasting model and an intelligent mining system. The system can forecast the buying and selling signs according to the prediction of future trends of stock market, and provide decision-making for stock investors. The simulation results of seven years of Shanghai composite index show that the return achieved by the system is about three times higher than that achieved by the buy-and-hold strategy. Accurate volatility forecasting is the core task in risk management in which various portfolios’ pricing, hedging, and option strategies are exercised. Roh proposes a hybrid model with neural network and various time series models for forecasting the volatility of stock price index from two viewpoints: deviation and direction (Roh, 2007). The results demonstrate the utility of the hybrid model for volatility forecasting. Mostafa proposed neural network-based mechanism to predict stock market movements in Kuwait using data for the period January 2001 to December 2003 (Mostafa, 2010). The author used two neural network architectures: multi-layer perception (MLP) neural networks and generalized regression neural networks to predict the Kuwait Stock Exchange (KSE) closing price movements. It had been demonstrated that due to their robustness and flexibility of modeling algorithms, neuro-computational models usually outperform traditional statistical techniques such as regression and ARIMA in forecasting price movements in stock exchanges. Kimoto et al applied neural networks on historical accounting data and used various macroeconomic parameters for the purpose of prediction of variations in stock returns (Kimoto et al, 1990). The authors presented a system for buying and selling timing prediction system for stocks in the Tokyo Stock Exchange. The learning algorithms proposed by the authors were demonstrated to produce accurate predictions of stock prices. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Leigh et al proposed the use of linear regression and simple neural network models for forecasting the stock market index in the New York Stock Exchange during the period 1981-1999 (Leigh et al, 2005). The authors proposed a mechanism to identify trading volume spikes through the use of a template matching technique based on statistical pattern recognition. The days that matched the condition signifying volume spikes, application of linear regression was done to model the future change in price using historical price and prime interest rate values. Hammad et al. demonstrated that artificial neural network (ANN) model can be trained to converge to an optimal solution while it maintains a very high level of precision in forecasting of stock prices (Hammad et al., 2009). The authors developed a back propagation algorithm for forecasting Jordanian stock prices using feed forward multi-layer neural network. MATLAB simulations conducted on seven Jordanian companies from service and manufacturing sectors produced extremely accurate predictions of stock prices. Dutta et al demonstrate the application of ANN models for forecasting Bombay Stock Exchange’s SENSEX weekly closing values for the period of January 2002 to December 2003 (Dutta et al, 2006). The authors developed two networks with three hidden layers. The first network was provided with the inputs of the weekly closing value, 52-week moving average of the weekly closing SENSEX values, 5-week moving average of the closing values, and the 10-week oscillator for the past 200 weeks. The second network was supplied with the inputs of weekly closing value, 52-week moving average of the weekly closing SENSEX values, 5-week moving average of the closing values and the 5-week volatility for the past 200 weeks. The performance of the two networks were evaluated by measuring their root mean square error values and the mean absolute error values in prediction of the weekly closing SENSEX values for the period January 2002 to December 2003. Tsai and Wang proposed a mechanism of combining ANN and decision tree-based approach to create a stock price forecasting model (Tsai & Wang, 2009). The experimental results demonstrated that the combined ANN and decision tree-based approach had higher forecast accuracy than the single ANN and the single decision tree-based approach. Tseng et al. deployed traditional time series decomposition (TSD), HoltWinters (H/W) models, Box-Jenkins (B/J) methodology and neural network- based approach on 50 randomly chosen stocks during September 1, 1998 - December 31, 2010 resulting in a total of 3105 observations for each company’s closing stock prices (Tseng et al, 2012). For hold-out period J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 or out-of-sample forecasts over 60 trading days, it has been observed that the forecasting errors are lower for B/J, H/W and normalized neural network model, while the errors are appreciably larger for time series decomposition and non-normalized neural network models. Moshiri and Cameron presented a Back Propagation Network (BPN) with econometric models to forecast inflation using (i) Box-Jenkins Autoregressive Integrated Moving Average (ARIMA) model, (ii) Vector Autoregressive (VAR) model and (iii) Bayesian Vector Autoregressive (BVAR) model (Moshiri, & Cameron, 2010). The authors compared each of these three approaches with a hybrid BPN model using the same set of variables. Forecasts were made over three different time horizons: one, three and twelve months. The performances of the models were compared using their root mean squared errors and mean absolute errors. It had been found that the hybrid BPN models outperformed the other approaches in many cases. Phua et al deployed ANNs with genetic algorithms for the purpose of predicting the stock prices in Singapore Stock Exchange (Phua et al, 2000). The result was promising with a forecast accuracy of 81% on the average. Hutichinson et al proposed a learning network-based non-parametric method for estimating the pricing formula of a derivative (Hutchinson et al, 1994). Some of the primary economic variables that influence the derivative price, e.g., the current fundamental asset price, the strike price, the time to maturity etc. were used as the inputs to the learning network. The derivative price was determined by the output into which the learning network mapped the inputs. The daily closing prices of S&P 500 futures and the options for the period from January 1987 to December 1991 were used as the training data set to construct the model. For the purpose of comparing the relative performance of various models, the authors used four methods: ordinary least squares, radial basis function networks, multilayer perceptron networks, and the projection pursuit. The non-parametric model was found to be more accurate in making derivative pricing forecasts than its parametric pricing counterpart. Chen et al. have proposed an approach for constructing a model for predicting the direction of return on the Taiwan Stock Exchange Index (Chen et al., 2003). The authors argued that trading strategies guided by forecasts of the direction of price movement are more effective and usually lead to higher profit. Probabilistic Neural Network (PNN) was used to forecast the direction of index return after it was trained using historical data. The authors applied their . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 forecasts to various index trading strategies, of which the performances were compared with those generated by the buy and hold strategy, and the investment strategies guided by the forecasts estimated by the random walk model and the parametric Generalized Method of Moments (GMMM) with Kalman filter. Basalto et al. applied a pair-wise clustering approach to the analysis of the Dow Jones Index companies, in order to identify similar temporal behavior of the traded stock prices (Basalto et al., 2005). The objective of the authors was to explore the dynamics which that govern the companies’ stock prices. A pairwise version of the chaotic map algorithm had been deployed that used the correlation coefficients between the financial time series to find the similarity measures for clustering the temporal patterns. The coupling interactions between the maps are considered as the functions the correlation coefficients. The resultant dynamics of such systems formed the clusters of companies that belong to different industrial branches. These clusters of companies can be gainfully exploited to optimize portfolio construction. Chen et al. studied how the seemingly chaotic behavior of stock markets could be very well represented using Local Linear Wavelet Neural Network (LLWNN) technique (Chen et al, 2005). The LLWNN was optimized by using Estimation of Distribution Algorithm (EDA). The objective was to predict the share price for the following trade day based on the opening, closing and maximum values of the stock price on a given day. The results showed that even in seemingly random fluctuations, there was an underlying deterministic feature that was directly enciphered in the opening, closing and the maximum values of the index of any day thereby making predictability quite possible. de Faria et al. proposed predictive framework for the principal index of the Brazilian stock market through ANN and adaptive exponential smoothing method (de Faria et al., 2009). The objective of the study was to compare the forecasting accuracies of both methods on the Brazilian stock index, with a particular focus on prediction of the sign of the market returns. Results showed that both methods were equally efficient in predicting the index returns. However, the neural networks outperformed the adaptive exponential smoothing method in forecasting of the market movement, with relative hit rates similar to ones found in other developed markets. Hanias et al. conducted a study to predict the daily stock exchange price index of the Athens Stock Exchange (ASE) using a neural network with back propagation (Hanias et al., J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Wu et al. proposed an ensemble model of SVM and ANNs for the purpose of predicting three stock index (Wu et al., 2008). The authors have compared the forecasting performance of the ensemble model with those of the SVM model and the ANN model. The results clearly indicated that the performance of the ensemble approach was superior to those of the other two models. Bentes et al. have investigated the long memory and volatility clustering for the S&P 500, NASDAQ 100 and Stoxx 50 indexes in order to compare the US and European markets (Bentes et al., 2008). The main objective of the authors were to compare two different perspectives: the traditional approach in which the authors have considered the GARCH (1, 1), IGARCH(1, 1) and FIGARCH (1, d, 1) specifications and the econophysics approach based on the concept of entropy. The authors had chosen three variants of this notion: the Shannon, Renyi and Tsallis measures. The results from both perspectives had shown nonlinear dynamics in the volatility of SP 500, NASDAQ 100 and Stoxx 50 indexes. Liao et al. investigated stock market investment issues on Taiwan stock market using a two-stage data mining approach (Liao et al., 2008). In the first stage, the authors have used the apriori algorithm to mine association rules and knowledge patterns about stock category association and possible stock category investment collections. After mining of association rules and knowledge patterns, the K-means clustering algorithm were used to identify the stock clusters in order to make a robust categorization of stocks in the Taiwan stock market. Several possible stock market portfolio alternatives under different situations had also been proposed. A number of work have been carried out by researchers using time series and fuzzy time series approaches for handling forecasting problems. Thenmozhi examined the nonlinear nature of the Bombay Stock Exchange time series using chaos theory (Thenmozhi, 2001). The study examined the Sensex returns time series from August 1980 to September 1997 and showed that the daily returns and weekly returns of the BSE sensex are characterized by nonlinearity and the time series is weakly chaotic. Fu et al. have proposed a method of representing a financial time series according to the importance of data points (Fu et al., 2007). Using the concepts of data point importance, the authors have constructed a tree data structure that supports incremental updating of the time series. The technique enables the user to present a time series in different levels of details and also facilitates multi-resolution dimensionality reduction of a large time series data. The J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 authors have presented several evaluation methods of data point importance, a novel method of time series updation, and two dimensionality reduction approaches. Extensive experimental results are also presented demonstrating the effectiveness of all propositions. Conclusion
Time series analysis of the index of several sectors of the economy of a country and the world can reveal several interesting points about the relationships of those sectors. A proper understanding of these relationships can prove critical in formulating several proactive economic policies in the current dynamic and extremely volatile economic conditions in the globe. As an illustration of the effectiveness of time series analysis in investigating the behavior and association among various sectorial index, in this paper, we proposed a time series decomposition-based approach for deeper understanding and analysis of two sectors of the Indian economy – the Indian IT sector and the Indian CG sector. We have hypothesized that the Indian IT sector, which is strongly coupled with the world economic story, should reveal a strong association with the DJIA index and the exchange rate of the US Dollar to the Indian Rupees, while the Indian CG sector, being essentially coupled with the Indian economic story, should expose a strong association with the NIFTY index. We also contended that although the Indian IT sector essentially reflects the world economic story, the blue chips stocks of the IT sector have strong impacts on the NIFTY index and hence the Indian IT sector should also have a strong association with the NIFTY index. However, the Indian CG sector being essentially India-centric only, we expect a poor association between the Indian CG sector index with the DJIA index and the US Dollar to Indian Rupee exchange rate. Using our proposed decomposition approach and several statistical tests on correlation and cross-correlation among different time series we have validated all our hypotheses. After carrying out detailed analyses and association tests on the time series of various sectors, and a comprehensive validation of our all hypotheses, we have built a number of linear models among time series which were found to have strong association between them. In other words, we have constructed linear regression models between the Indian IT sector index with the DJIA index, the exchange rate of the US Dollar to the Indian Rupee and the NIFTY index. For the Indian CG sector, we have built a linear model with the NIFTY index. All the linear models of regression are built using suitably chosen training data set and their robustness and efficacies are tested using appropriate . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131 test data set. Forecast errors are computed for each month of the period of forecast so that the effectiveness of each model can be numerically evaluated. The approach and methodology proposed in this paper are absolutely generic and can be applied to any sector of the world economy. Moreover, the results obtained using this approach can also be extremely useful for portfolio construction of stocks. By performing analysis on time series of several sectors and studying their seasonality characteristics, portfolio managers and individual investors can very effectively take decisions about buy/sell of stocks and appropriate timing. For speculative gains, the sectors exhibiting presence of dominant random components in their time series may be targeted.
8. References
Basalto, N., Bellotti, R., De Carlo, F., Facchi, P., & Pascazio, S. (2005). Clustering stock market companies via chaotic map synchronization.
Physica A , 345, 196-206. Bentes, S. R., Menezes, R., & Mendes, D. A. (2008). Long memory and volatility clustering: is the empirical evidence consistent across stock markets?
Physica A: Statistical Mechanics and its Applications , 387(15), 3826-3830. Chen, A.-S., Leung, M. T. & Daouk, H. (2003). Application of neural networks to an emerging financial market: forecasting and trading the Taiwan stock index.
Computers & Operations Research , 30(6), 901– 923. Chen, Y., Dong, X. & Zhao, Y. (2005). Stock index modeling using EDA based local linear wavelet neural network.
Proceedings of International Conference on Neural Networks and Brain , Beijing, Vol 3, 1646–1650. Coghlan, A. (2015).
A Little Book of R for Time Series , Release 02. Available online: https://media.readthedocs.org/pdf/a-little-book-of-r-for-time-series/latest/a-little-book-of-r-for-time-series.pdf. (last accessed on: May 9, 2016). De Faria, E. L, Albuquerque, M. P., Gonzalez, J., Cavalcante, J. T. P. & Albuquerque, M. P. (2009). Predicting the Brazilian stock market through neural networks and adaptive exponential smoothing methods.
Expert Systems with Applications , 36(10), 12506 – 12509. Dutta, G., Jha, P., Laha, A. & Mohan, N. (2006). Artificial neural network models for forecasting stock price index in the Bombay stock exchange.
Journal of Emerging Market Finance , 5, 283-295. Fu, T-C, Chung, F-L., Luk, R., & Ng, C-M, (2008). Representing financial time series based on data point importance.
Engineering Applications of Artificial Intelligence , 21(2), 277-300. Hammad, A. A., Ali, S. M. & Hall, E. L. (2007). Forecasting the Jordanian stock price using artificial neural network.
Intelligent Engineering Systems through Artificial Neural Networks , ASME Press, USA. Hanias, M., Curtis, P. & Thalassinos, J. (2012). Time series prediction with neural networks for the Athens stock exchange indicator.
European Research Studies,
Neural Networks , 2(5), 359-366. J. Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Hutchinson, J. M., Lo, A. W., & Poggio, T. (1994). A nonparametric approach to pricing and hedging derivative securities via learning networks.
Journal of Finance , 49(3), 851-889. Ihaka, R. & Gentleman, R. (1996). A language for data analysis and graphics.
Journal of Computational and Graphical Statistics , 5(3), 299-314. Kimoto, T., Asakawa, K., Yoda, M. & Takeoka, M. (1990). Stock market prediction system with modular neural networks.
Proceedings of the IEEE International Conference on
Neural Networks , 1-16. Leigh, W., Hightower, R. & Modani, N. (2005). Forecasting the New York stock exchange composite index with past price and interest rate on condition of volume spike.
Expert Systems with Applications , 28, 1-8. Liao, S-H., Ho, H-H., Lin, H-W. (2008). Mining stock category association and cluster on Taiwan stock market.
Expert System with Applications , 35(2008), 19-29. Moshiri, S. & Cameron, N. (2010). Neural network versus econometric models in forecasting inflation.
Journal of Forecasting , 19, 201-217. Mostafa, M. (2010). Forecasting stock exchange movements using neural networks: empirical evidence from Kuwait.
Expert Systems with Application , 37, 6302-6309. Ning, B., Wu, J., Peng, H. & Zhao, J. (2009). Using chaotic neural network to forecast stock index.
Advances in Neural
Networks,
Lecture Notes in Computer Science (LNCS) , 5551, 870–876. Phua, P. K. H., Ming, D., & Lin, W. (2000). Neural network with genetic algorithms for stocks prediction.
Proceedings of the 5th Conference of the Association of Asian-Pacific Operations Research Societies , Singapore. Roh, T. H. (2007). Forecasting the volatility of stock price index.
Expert Systems with Applications,
Proceedings of the 2 nd National Conference on Advances in Business Research and Management Practices (ABRMP’16) , Kolkata, India, January 8 -9, 2016. Available online at: https://arxiv.org/abs/1601.02407 (last accessed on: May 31, 2016) Sen, J. & Datta Chaudhuri, T. (2016b). A framework for predictive analysis of stock market index – a study of the Indian auto sector.
Calcutta Business School (CBS) Journal of Management Practices , 2(2), 1 – 19. Available online: https://arxiv.org/abs/1604.04044 (last accessed on May 31, 2016) Sen, J. & Datta Chaudhuri, T. (2016c). An alternative framework for time series decomposition and forecasting and its relevance for portfolio choice: a comparative study of the Indian consumer durable and small cap sectors.
Journal of Economics Library , 3(2). Accepted for publication. Available online: https://arxiv.org/abs/1605.03930 (last accessed on May 31, 2016) Shen, J., Fan, H. & Chang, S. (2007). Stock index prediction based on adaptive training and pruning algorithm.
Advances in Neural Networks , Lecture Notes in Computer Science , Vol 4492, 457–464. Shumway, R. H. & Stoffer. D. S. (2011).
Time Series Analysis and Its Applications . Springer-Verlag, New York, USA. Thenmozhi, M. (2006). Forecasting stock index numbers using neural networks.
Delhi Business Review , 7(2), 59-69. Tsai, C.-F. & Wang, S.-P. (2009). Stock price forecasting by hybrid machine learning techniques.
Proceedings of International Multi Conference of Engineers and Computer Scientists , 1. Tseng, K-C., Kwon, O., & Tjung, L. C. (2012). Time series and neural network forecast of daily stock prices.
Investment Management and Financial Innovations , 9(1), 32-54. . Sen et al. / Journal of Insurance and Financial Management, Vol. 1, Issue 4 (2016) 68-131
Wang, W. & Nie, S. (2008). The performance of several combining forecasts for stock index.
International Seminar on Future Information Technology and Management
Engineering , Leicestershire, UK, 450– 455. Wu, Q., Chen, Y. & Liu, Q. Z. (2008). Ensemble model of intelligent paradigms for stock market forecasting.
Proceedings of the IEEE 1 st International Workshop on Knowledge Discovery and Data Mining , 205 – 208, Washington, DC, USA. Zhang, D., Jiang, Q., & Li, X. (2007). Application of neural networks in financial data mining.
International Journal of Computer, Electrical, Automation, and Information Engineering,