Pankaj Agrrawal

What is Wrong with this Picture? A Problem with Comparative Return Plots on Finance Websites and a Bias Against Income-Generating Assets

This paper brings to light and discusses a systemic issue in the calculation and display of relative return information as currently seen on some of the most prominent finance websites; income-generating events such as dividends and interest are not included in relative return calculations and all comparative return graphics. The resulting ranking of the securities, based on such incomplete returns, is essentially meaningless from a total return perspective, yet they are being served to millions of investors every day. This could lead to the formation of a possible availability heuristic and an optical bias against fixed-income and other income generating assets. This problem has gone unnoticed for many years with no discussion of the topic either in the academic or practitioner press. The ready availability of such unclear or inaccurate information from sources generally perceived to be credible can, in this age of do-it-yourself portfolio management, have serious and damaging financial consequences to the unsuspecting investor. The paper also shows the effect of this return differential on the calculation of the asset correlation matrices and the subsequent effect on the resulting asset-weight vectors that are used to generate Markowitz style mean-variance portfolios. The visual discrepancies are then supported by the application of the Gibbons, Ross and Shanken [1989] W-test for portfolio efficiency. The authors’ proposed correction, based on elementary finance, fixes the problem.

An automation algorithm for harvesting capital market information from the web

Purpose - The purpose of this paper is to develop an algorithm to harvest user specified information on finance portals and compile it into machine-readable datasets for quantitative analysis. Design/methodology/approach - The Visual Basic macro language in Microsoft Excel is applied to develop code that is not constrained by the single-query function of Excel. The core of the algorithm is built around the splitting of the URL connector line and the placement of a continuously updating variable into which are looped as many tickers as there are in the input list. The output is then written to non-overlapping cells. Findings - Numerical information placed on major finance websites can be harvested into structured machine-readable datasets by applying this algorithm. Research limitations/implications - One significant change in Microsoft Excel 2007 is that the worksheet is expanded from 224 to 234 cells, or to be more specific, from 256 (IV) columns × 65,536 rows (28 × 216) to 16,384 (XFD) × 1,048,576 (214 × 220). These new limits while allowing for a larger number of tickers, still constrain a single worksheet to 16,384 columns. For five fields per ticker that translates into roughly 3,200 ticker symbols. Practical implications - The algorithm extends user accessibility to websites that do not provide the facility of simultaneous downloading of information on multiple stock tickers. Furthermore, the procedure automates the downloading of multiple pieces of information (fields) and entire tables per ticker (record). Originality/value - An exhaustive literature search did not find any paper that discusses a multiple ticker algorithm for web harvesting.

Using Index ETFs for Multi-Asset-Class Investing: Shifting the Efficient Frontier Up

This article provides evidence and analysis to show that a MAC (multi-asset-class) diversified portfolio performed well in mean–variance space and under varying market conditions, including the very adverse 2008 market crash. The portfolio also delivered during the two bull phases in the full period over which asset history existed. The construction of the covariance matrix for the efficient frontiers was independent of any return estimates or dynamic volatility-switching mechanisms. To abstract from hindsight bias, a 1/N equal-weighted portfolio was constructed and tested, consistent with some literature—it may still be the best alternative. In any case, the minimum-variance portfolios and the 1/N portfolio far exceeded the Sharpe ratio of the capitalization-weighted Russell 1000 equity index. The efficiency gains are potentially attributed to a lower overlap of the return-generating vectors, something that is not possible, to that extent, in an all-equity portfolio, irrespective of the extent of diversification in the non-negative space. Toward that, a scalar construct of overall dependencies called generalized variance is used as a measure of the overall spread within the covariance matrix. Two actual efficient frontiers are built with return data over the full length of the study and are tested for portfolio efficiencies. Finally, with the objective of making such alternate asset and risk-allocation processes available to a wider set of investors, the portfolio components chosen to represent the low-correlation asset classes were among the most liquid index ETFs available on U.S. exchanges.

The Dispersion of ETF Betas on Financial Websites

This article documents significant dispersion in the beta estimates of exchange-traded funds as available on some leading financial websites. To the best of the authors’ knowledge, this is the first systematic study of the dispersion of betas as seen on major finance websites. Almost 40 million visitors access these websites per month. The authors find that leading sites such as Yahoo! Finance, MSNMoney, Morningstar, and Google Finance display betas that significantly (but not intentionally) misrepresent actual levels of systematic risk. These errors could impact the portfolio design of an investor, leading to unintended outcomes. Additionally, the authors identify the primary reason for the significant variance in beta estimates. The explanation is surprising and not rooted in the traditionally discussed differences that are attributed to interval-window length mismatch or varying frequency of security returns. An implication of the findings is that there is no substitute for verification and cross-validation of financial information, even if it is supposedly coming from well-established sources or so-called “black boxes.”

Seasonality in Stock and Bond ETFs (2001-2014): The Months are Getting Mixed Up But Santa Delivers on Time

This article examines the current state of seasonality in returns using a set of ten highly liquid exchange-traded funds (ETFs). Our analysis extends beyond the traditional stock market framework to also include bond, real estate, and gold assets, in the same study. Additionally, this use of ETFs is a new approach compared with existing seasonality literature. Four well-known effects are researched – the January effect, the Halloween effect (“Sell in May and Go Away”), the Mark Twain effect, and the Santa Claus rally. The results are mixed. Some seasonality effects seem to have weakened while others have remained intact or even strengthened. This might be the result of improved market efficiency, arbitrage activity, or high frequency trading (HFT). The persistence of some of the effects is somewhat puzzling against the backdrop of the efficient market hypothesis (Fama [1970]) but could be rationalized via differential investor sentiment responses (Waggle and Agrrawal [2015]). The article also provides reference tables that include probabilities and averages for each month and for each effect. The Altman-Wald and Friedman tests are utilized for statistical significance, given the relatively short return histories for ETFs. These could be utilized by a trader as a tactical overlay on top of a longer term strategic allocation. Finally, we introduce the reader to the bond based “Safety in Summer” effect, as an additional calendar effect, to be further researched in the years to come.

Investor Sentiment and Short-Term Returns for Size-Adjusted Value and Growth Portfolios

We examine the sentiment levels of individual investors relative to subsequent short-term market returns for 1992–2010. We find that sentiment, proxied by percentage of investors who are “bullish” on the market, is significantly negatively related to the subsequent three- and six-month performance of the market. The negative relationship is consistent with the contrarian notion of sentiment. In other words, high (low) levels of bullishness tend to be followed by subsequent low (high) returns. This is true even with the inclusion of standard control explanatory variables (Fama-French [1993]). While the significant results hold for the overall market, they are clearly driven by growth, rather than value stocks. Contrary to some earlier studies, we also note significant explanatory power for sentiment when looking at returns of small-, mid-, and large-cap growth stocks. We also noted that the long-term moving average of monthly bullishness increased from 33.3% to 39.0% over the last 18 years. In our study period, about 5% of the total sentiment observations are above 56% (very bullish) and about 5% are below 27% (quite bearish). Finally, we find some strength in the lagged autocorrelation structure for the sentiment variable that lasts for just about three to nine months.

An Intertemporal Study of ETF Liquidityand Underlying Factor Transition, 2009–2014

This article seeks to determine the migration of exchange-traded fund (ETF) liquidity and its factor constituents in the U.S. market over time, with the ultimate goal of making the ETF market more efficient and transparent. Using a set of factors commonly thought to impact liquidity, the authors develop a four-factor liquidity scoring algorithm (A-C liquidity score), extending the 2009 study by Agrrawal and Clark. The most liquid ETFs typically have lower bid–ask spreads, higher market capitalizations, lower expense ratios, and higher average trading volumes. The transition of ETF liquidity over the 2009–2014 period indicates that there is liquidity persistence and factor strengthening across all variables. Bid–ask spreads and expense ratios have compressed, which is a good trend for investors. Asset size and trading volume have gone up, which can be indicative of greater market participation and possibly high-frequency trading in the ETF environment. From 2009 to 2014, the ETF asset class has witnessed a phenomenal 34.9% annual growth rate, compared with a 19.8% growth rate in the Russell 3000 (the aggregate market capitalization for the ETF dataset increased from

Suicides as a response to adverse market sentiment (1980-2016)

438 billion in 2009 to

Excess Capacity in Banking: Fact or Fiction?

2,112 billion in 2014). The ETF ecosystem seems to be thriving as the new asset class of choice among both institutional and informed retail investors. However, there are two findings of concern. First, the overall beta of the ETFs in the study seems to have transitioned over time and moved up more than 9%. This may be due to increased cross-asset correlations, sector intertwining, global market cointegration, overlapping core constituents in ETFs, or simply a systematic upward drift in overall market risk premium; this limits portfolio diversification benefits. Second, over the five-year period, 25.7% of ETFs in the initial 622 ETF dataset (2009) were liquidated. This is an additional risk that investors and regulators have to keep in mind when evaluating ETFs. The cost of poor liquidity, overlooked by some retail investors, is often in the form of shallow market depth, high expense ratios, and a range of trade execution difficulties, besides potentially high tracking errors with the associated benchmark index. In conclusion, although ETF liquidity has grown significantly over time for most assets, there is some natural pruning, with pockets of concern. Overall, ETFs have rapidly become a viable and efficient instrument in the asset management landscape.

The Stock-REIT Relationship and Optimal Asset Allocations

Financial crises inflict significant human as well as economic hardship. This paper focuses on the human fallout of capital market stress. Financial stress-induced behavioral changes can manifest in higher suicide and murder-suicide rates. We find that these rates also correlate with the Gross Domestic Product (GDP) growth rate (negatively associated; a -0.25% drop [in the rate of change in annual suicides for a +1% change in the independent variable]), unemployment rate (positive link; 0.298% increase), inflation rate (positive link; 0.169% increase in suicide rate levels) and stock market returns adjusted for the risk-free T-Bill rate (negative link; -0.047% drop). Suicides tend to rise during periods of economic turmoil, such as the recent Great Recession of 2008. An analysis of Centers for Disease Control and Prevention (CDC) data of more than 2 million non-natural deaths in the US since 1980 reveals a positive correlation with unemployment levels. We find that suicides and murder-suicides associated with adverse market sentiment lag the initial stressor by up to two years, thus opening a policy window for government/public health intervention to reduce these negative outcomes. Both our models explain about 73 to 76% of the variance in suicide rates and rate of change in suicide rates, and deploy a total of four widely available independent variables (lagged and/or transformed). The results are invariant to the inclusion/exclusion of 2008 data over the 1980–2016 time series, the period of our study. The disconnect between rational decision making, induced by cognitive dissonance and severe financial stress can lead to suboptimal outcomes, not only in the area of investing, but in a direct loss of human capital. No economic system can afford such losses. Finance journal articles focus on monetary alpha, which is the return on a portfolio in excess of the benchmark; we think it is important to be aware of the loss of human capital as a consequence of market instability. This study makes one such an attempt.