Diagnosis of systemic risk and contagion across financial sectors
DDiagnosis of systemic risk and contagion across financial sectors
Sayuj Choudhari and Richard Zhu Adlai E. Stevenson High School Department of Computational and Applied Mathematics, University of ChicagoDecember 2020 a r X i v : . [ ec on . GN ] J a n ontents Introduction
As is expected with hindsight, there were obvious signs that significant systemic risk had been pent-up in theeconomy prior to the Great Financial Crisis (GFC) of 2007-2009. Deregulation beginning nearly a decadeprior enabled many financial institutions to take on massive leverage. From May 2000 to June 2003, theFederal Reserve cut interest rates from 6.5% to 1% over in response to the crash of the dot-com bubble,multiple accounting scandals, and the terrorist attacks of September 11th. While this provided a shortterm fix, cut interest rates replaced the dot-com bubble with a real estate bubble. Lenders originated loansto many people financially unfit to buy homes, passing them onto the burgeoning secondary market fororiginating and distributing subprime loans (Investopedia, 2020).The repeal of the Glass-Steagall Act in 1999 loosened the rules on the abilities of banks and brokerages,allowing them to both take deposits and make loans, as well as underwrite and sell securities. Rickards(2012) explains the danger of this: these institutions could originate loans, repackage the cash flows fromthe loans as securities, and pass off these mortgage-backed securities to other buyers. Because these novelsecurities paid huge fees to the originators, more and more institutions decided to get in on the game. TheSEC also lowered net asset requirements for Goldman Sachs, Merrill Lynch, Lehman Brothers, Bear Stearns,and Morgan Stanley, which enabled these institutions to increase leverage massively. Repackaging MBSsbecame a shell game, creating a scheme where trillions of dollars of securities held on balance sheets couldbe tethered to the performance of mere billions of mortgages.In normal times, it is assumed that financial institutions operating in non-overlapping sectors have com-plementary and distinct outcomes, typically reflected in mostly uncorrelated outcomes and asset returns.Such is the reasoning behind common ”free lunches” to be had in investing, like diversifying assets acrossequity and bond sectors. Unfortunately, the recurrence of crises like the GFC demonstrate that such con-venient assumptions often break down, with dramatic consequences for all financial actors. The emergenceof so-called systemic risk (exemplified by failure in one part of a system spreading to ostensibly unrelatedparts of the system) is thus far clearly evinced in the deregulation narrative. But is it possible to diagnoseand quantify the emergence of systemic risk in financial systems?In this study, we focus on two previously-documented measures of systemic risk that require only easily-available time series data (eg monthly asset returns): cross-correlation and principal component analysis.We apply these tests to daily and monthly returns on hedge fund indexes and broad-based market indexes,and discuss their results. We hope that a frank discussion of these simple, non-parametric measures can helpinform legislators, lawmakers, and financial actors of potential crises looming on the horizon.Section 2 provides a review of literature on previous research and methods used to analyze and measure3ystemic risk. The measures and methods used in this study are described in Section 3. The data tablesand plots of the results from the systemic risk measurement analyses conducted is presented in Section 4, aswell as a brief explanation of the meaning of the data found. The analysis of this data with the conclusionsof this study are given in Section 5.
Following the work of Andrew Lo (2010) we use the definitions of systemic risk and the crises it causes thatare proposed by De Bandt and Hartmann (2000):A systemic crisis can be defined as a systemic event that affects a considerable number of financialinstitutions or markets in a strong sense, thereby severely impairing the general well-functioning of thefinancial system. While the “special” character of banks plays a major role, we stress that systemic risk goesbeyond the traditional view of single banks’ vulnerability to depositor runs. At the heart of the conceptis the notion of “contagion”, a particularly strong propagation of failures from one institution, market orsystem to another.We use the the requirements for systemic risk measure described by Brunnermeier et al. (2009) which is:A systemic risk measure should identify the risk on the system by individually systemic institutions,which are so interconnected and large that they can cause negative risk spillover effects on others, as well asby institutions which are systemic as part of a herd.The statistical tests we conduct in this research follow those definitions for the description and measure-ment of systemic risk. The importance of investigating the connectedness of financial institutions/ sectorsin the study of systemic risk, which is done in this study, is shown in Allen (2001) where the interlinkagesbetween financial institutions and its effects are identified. These evaluations of interconnectedness withinthe finance world are further studied by many other researchers. Barunik (2018) evaluates the dynamicsof U.S. financial institutions for interlinkages on the short and long term and their varying effects on theeconomy. The book “Financial and Macroeconomic Connectedness: A Network Approach to Measurementand Monitoring” by Francis Diebold and Kamil Yilmaz 2015 also develops strategies to identify relationsbetween financial institutions to create a deep understanding of the global financial network and the causesand effects of financial interlinkages. Aikman (2017) proposes the use of many types of economic statisticsfor various financial institutions to map out areas of vulnerability within the U.S. economy.4eeing that the interconnectedness between financial institutions was becoming an increasingly populararea of study for systemic risk, we now had to look into a specific financial institution/ sector to focus our testson. De Bandt and Hartmann’s definition of systemic risk emphasized that banks, although seemingly playinga significant role in market crashes and systemic failures, are not the only cause of such crises as systemicrisk goes beyond a singular institution or system making it necessary to consider other parts of the marketas well. Thus, we looked into what financial sectors may have played important roles in crises and fromwhich of these major sectors systemic risk may have stemmed from. Multiple sources of literature were foundinvestigating or citing hedge funds as an early indicator of systemic risk in the recent 2008 market crash. Lo(2008) describes and evaluates the relation between systemic risk and hedge funds during that time period,and furthered his research into showing possible “early warning signs” seen within the hedge fund sector ofthe coming market crash. Systemic risk by hedge funds imposed on the banking sector (commonly believedas the key role in systemic failures) is studied by Chan (2005), demonstrating the possibility of systemic riskwithin banking in 2008 to have stemmed from the hedge fund sector. Furthermore, the effect of hedge fundson counterparty credit risk management (CCRM, systemic risk and market crash prevention measures) isexamined in Kambhu, Schuermann, and Stiroh (2007), where the different structures and operations of hedgefunds are shown to challenge CCRM and increase systemic risk. These works along with many others, ledus to focus our testing of systemic risk on the hedge fund sector and investigate the statistical data betweenhedge funds with other sectors as well as within different hedge funds themselves.Our use of Principal Component Analysis (PCA) to measure systemic risk is closely based off of Lo (2010)in which hedge funds, banks, brokerages, and insurers were tested for systemic risk using the return dataof the indexes of the mentioned financial sectors. Similar work is also seen in Kritzman, Li, Page, Rigobon(2011) where principal components are used to measure the connectedness of institutions within a marketas well as the market fragility. This group of researchers extends their work in 2012 to show the use of PCAto evaluate the risk specific entities have on the market, how vulnerable specific entities are to systemicrisk. The use of cross-correlations to analyze systemic risk in this study is also complementary to that ofLo (2001) in which the cross-correlation coefficients between 2 types of financial data are used to identifypossible systemic risk within the market. The single comparison used in cross-correlations, which cannot bedone in PCA, is helpful in creating more highly defined mappings of systemic risk and connectedness withinthe financial markets, as they allow for the analysis of a singular sector/ institution separately against variousothers. Chiang, Jeon, and Li (2007) also implemented the methods of correlations on the Asian market todiscover different financial phases of the asian market crash. Also along the same lines, Getmansky, Lo, andMakarov (2004) examine serial correlation within hedge fund returns and the high correlation levels beinglikely explained by the illiquidity exposure within the hedge fund industry, which may have initiated systemic5isk. The timing and results of this work showing illiquidity exposure risk within the hedge fund industryonly a few years before the financial crisis of 2008 further motivated us to emphasize our work towards themovements of hedge funds.
Principal component analysis (hereafter, known as PCA) involves the eigenvalue decomposition betweendata sets/ components, which calculates the amount of variation brought by each data set/ component (seeMuirhead, 1982 for an exposition of PCA). In context, when most of the variation is accounted for in thefirst few components this would mean that the components are more correlated as there is less variationamong them and their movements are similar. Therefore, when applying PCA to finance, each componentcan be a time-series return data on an asset, and detect increased correlations among assets over certaintime periods, which in other words is measuring systemic risk and the exposure each asset has on the other.In this study, PCA is conducted on time-series data for hedge funds with the S&P 500 to detect possiblesystemic risk between hedge funds and the overall US economy, as well as just on various hedge fund sectorsthemselves to detect possible systemic risk within the hedge fund industry itself.The first step in the PCA of this study is to create a returns matrix of the data (hereafter, known as R ),where each row is an asset (In this study: LAB hedgefund index, banks, brokerages, and insurers) and eachcolumn is each row (asset) is return data at a certain step in the time-series, R ( n , k ). The mean return ofeach asset is then subtracted from each respective row creating the (insert name of matrix type) matrix of R (Matrix M ): M n,k = R ,k − R ,k R ,k − R ,k ... R n − ,k − R n − ,k R n,k − R n,k Matrix M, now represents the variance of each asset is return data from the mean of the returns. Thus,when M is multiplied by the transposed version of itself, M T , the covariance matrix Σ is given:6 ar [ R t ] ≡ Σ = MM T = M ,k · M Tk, . . . M ,k · M Tk,n ... ... M n,k · M Tk, . . . M n,k · M Tk,n
As seen above, each component of Σ represents the correlation between the two assets that are representedin that location. For example, the component (n,k) in Σ represents the value n k + n k + .... + n M k M where n t and k t is the variance of return for assets n and k respectively over a time step. Looking back atthe calculation of that component ( n, k ) it is seen that the component can indicate the typical movementsbetween the two assets. Considering that this calculation is done for all components of Σ, Σ is representativeof all the movements between all the assets. Thus, when we take the eigenvalue decomposition of Σ we canobserve the weightage of each assets movements, this can be expressed as: Var [ R t ] ≡ Σ =
QΘQ (cid:48) , Θ = θ . . . θ . . . θ N In the above equation, matrix Θ would have the eigenvalues of Σ along it’s diagonal along with Matrix Q containing the corresponding eigenvectors. Thus, we get a set of eigenvalues, where each eigenvalue findsthe total variation accounted for in each principal component which is essentially reflecting the importanceof each component is movements for all assets analyzed. Once these eigenvalues are then normalized to asum of 1, we are given the fractional eigenvalues corresponding with each principal component which allowus to analyze where the variation of the assets is movements is accounted for. One issue that arises with theuse of PCA is that the covariance matrix Σ must be estimated: (cid:98) Σ ≡ T − J T (cid:88) t =1 ( R t − R )( R t − R ) (cid:48) is singular when the number of assets J compared as larger than the amount of time-series data points T available for each asset. Thus, the number of assets we compare is limited to only hedge funds, banks,brokerages, and insurers (Lo, 2010), and the windows we use for each PCA calculation is relatively large (30datapoints) to maximize degrees of freedom and create more accurate results.So, using Principal Component Analysis, if the fractional variation of the first or first few eigenvalueswas very high and quickly approached 1, then the movements of all assets compared are very correlated asmost variation is accounted for in only one or a few principal component(s) and pose a possibility of having7ystemic risk. On the other hand if the fractional variation of the assets does not quickly approach the noisearound 1 within the first few components, then the variation is spread across all assets and movements aretherefore uncorrelated and do not indicate systemic risk.In this study, the fractional variation of the first component is calculated over a time period of (TBD) asit is best representative of the correlation between all assets whereas after 3 components are considered thefractional variation goes into the noise approaching 1. Mapping out the data of each PCA on a (TBD-length)time period over the entire time-series of data, we can detect the time-periods when fractional variation ishighest and systemic risk is likely present. In order to determine the influence or exposure of asset movements on each other, cross correlations cancalculate coefficients that determine the extent as to how much one asset is movements are reflected inanother. The reason as to why similar movements over a period of time are a result of correlations ratherthan chance can be derived from the martingale model: the idea that all information in an informationallyefficient environment is accounted for in an assets price making future prices unpredictable and random(Samuelson, 1965). However, this informationally efficient environment required in the martingale modelis never fully achieved as the market can never be perfectly efficient, and therefore predictable points ormarket friction can occur causing correlation/ exposure between assets. This friction is an example of assetsbeing influenced/ exposed to others as the movements of one asset are creating the movements of another,rather than a result of chance. Furthermore, in a time period of systemic risk, these correlations/ exposuresbetween assets must be increasing as the martingale model fails and market friction increases (Lo, 2010).With this direct relationship between systemic risk and correlation between assets, testing cross-correlationsto find levels of correlation between assets is a possible way to find systemic risk levels between those assetsas well. Cross-correlation also allows comparison of one asset is movements to a lagged version of the other.This becomes helpful in the case that an asset is movements are a lagged reflection of another, as the effectsof market friction and systemic risk may take a certain time period to set in.The cross-correlation coefficient is the variable within a cross-correlation calculation that indicates thelevels of correlation between two assets with a possible time lag ( l ). The data necessary for this calculationis a time-series of return data on an asset with a constant time-step. The first step is to create a data setfor each asset that is the variance of that asset is actual returns over a time period from the mean returnsover that time period: 8 ar [ A ] = A − AVar [ B ] = B − B Using these two new data lists for the two assets one of the asset lists is converted into a convolutedmatrix and multiplied by the data list of the second asset. This finds the sum of all covariances between thetwo assets at any time lag in the range of the time period tested for l ∈ [0 , k) where k is the length of thetime period: C = A ∗ B = A . . . A A ... ... A A . . . A . . . A A k − ... . . . A A A k A k − ... A A k . . . A k − ...0 0 . . . A k − A k − ... ... A k A k − . . . A k B B ... B k − B k As seen in the equation above, a vector of length 2 k - 1 is created, with the center component being thesum of covariances at a time lag of 0 (e.g. considering asset A and asset B the center of this vector wouldbe A B + A B + ... + A k B k where k is the length of the time period tested over). Therefore, all vectorcomponents ahead of that center component are truncated to a new vector that is the same length as thetime period D = [ C , C , ... , C k − , C k − ] and contains all covariance sums between the two assets for l ∈ [0 , k ). Now considering that the covariance sum with l = 0 was composed of covariances on all time stepsin the time period, while the covariance sum with l = k − r ) at each time lag: r = D [1 , ,...,k − ,k ] ∗ σ ( V ar [ A ] ,V ar [ B ]) The Hedge Fund data used in this study comes from the Credit Suisse Liquid Alternative Beta Index(hereafter, known as “LAB Index”) and contains daily pricing information and can therefore be used tofind return data over any period of time from January 1998 to June 2020. It consists of liquid securitiesand aims to reflect the returns of the overall hedge fund industry strategies that use liquid and tradableinstruments. The LAB Index develops a measure for returns on the entire hedge fund industry throughincluding hedge fund data from 5 different Credit Suisse Hedge Fund Indices that differ in strategy: theEvent Driven Liquid Index (hereafter, known as “EDL Index”), Global Strategies Liquid Index (hereafter,known as “GSL Index”), Long/ Short Liquid Index (hereafter, known as “LSL Index”), Managed FuturesLiquid Index (hereafter, known as “MFL Index”), and Merger Arbitrage Liquid Index (hereafter, known as“MAL Index”). The LAB Index takes into account the different weights each of these indices have in theoverall hedge fund industry. This weighting is reflective of each strategy’s respective weight in the CreditSuisse Hedge Fund Index.This research also includes the separate pricing data for each of the 5 LAB sub-indices. This separatedata for each hedge fund strategy helps to run tests for systemic risk within the hedge fund industry itselfthrough providing individual data on these 5 hedge fund sectors. The hedge fund sectors that are part ofthese indices are all susceptible to systemic risk from the other sectors as they all are part of the LAB index,which as known earlier, contains liquid hedge fund strategies that are exposed to economic movements/events occurring. Thus, systemic risk within the hedge fund industry can be seen through these 5 indicesand should be measured. Furthermore, it is necessary to establish a definition for each of these indices as10o help provide reasoning for the results of the systemic measurement tests conducted in Section 5. Asaccording to Credit Suisse, the following are the definitions for each of the indices /indice strategies:EDL Index: Develops trading strategies to create returns and solid out of global events (e.g. elections,corporate situations/ activities, mergers and acquisitions deals, etc.)GSL Index: Reflects the return of all remaining hedge fund strategies not defined as Long/Short or EventDrivenLSL Index: Strategy that takes a long term position on underpriced equities and also selling short equitiesthat are expected to decline in valueMFL Index: Uses a pre-defined quantitative methodology to invest in a range of asset classes including:equities, fixed income, commodities and currenciesMAL Index: Uses using a pre-defined quantitative methodology to invest in a liquid, diversified andbroadly representative set of announced merger deals
The S&P 500 data is obtained from investing.com and contains the daily pricing and return values fromJanuary 2006 to June 2020. Due to this data only containing returns from 2006 onwards, while the hedgefund data from Section 4.2 goes back to 1998, the hedge fund data for days that do not correspond withthe S&P 500 data is truncated during the tests that compare the two different financial sectors. The S&P500 is an asset-weighted index containing the top 500 companies in the US market, and is used as a generalevaluation for the US economy as a whole. Similar to the hedge fund data, the daily information provided byinvesting.com allows to calculate return values over any time period within the range of the data, meaningmonthly data can be created in cases that the second data set compared with contains monthly data. Datathat represents the entire US economy and its movements is relevant in this study as it provides the meansto measure systemic risk of the hedge fund industry on the overall US economy. More specifically it alsoallows testing each of the LAB Index is 5 subindices for possible systemic risk with the US economy. Fromthose tests, we can determine which hedge fund strategies might be susceptible to or creating systemic riskwithin an economy. Overall, the S&P 500 data provides a gauge of the US economy which is important totest with the hedge fund data as it can evaluate the relations and movements between hedge funds and therest of the financial/ corporate world.
The bank data is obtained from the Wharton School of Economics is WRDS service with the CRSP dailystock price database and contains daily pricing and return information from February 1990 to June 2020 of11anks with SIC codes from 6000-6199. The return data is then used along with the market capitalization ofeach firm to create a market-cap weighted index of returns for the banks. This banking index will serve asa measurement of the entire banking sector. The data is truncated for the same date range from January2006 to June 2020 as mentioned in 4.2. Similar to all other data in this study, the daily returns from thebanking index can be used to create return data for any length time period within the time range. Thisbanking data is important as it provides the banking sector component of the financial industry that whencombined with other components from this study (hedge funds, brokerages, and insurances) can be testedfor systemic risk within the entire economy using PCA (statistical test explained in 3.2).
The brokerage data is obtained from the Wharton School of Economics is WRDS service with the CRSPdaily stock price database and contains daily pricing and return information from February 1990 to June2020 of brokerages with SIC codes from 6200-6299. The return data is then used along with the marketcapitalization of each firm to create a market-cap weighted index of returns for the banks. This bankingindex will serve as a measurement of the entire brokerage sector. The data is truncated for the same daterange from January 2006 to June 2020 as mentioned in 4.2. Similar to all other data in this study, the dailyreturns from the brokerage index can be used to create return data for any length time period within thetime range. This brokerage data is important as it provides the brokerage sector component of the financialindustry that when combined with other components from this study (hedge funds, banks, and insurances)can be tested for systemic risk within the entire economy using PCA (statistical test explained in 3.2).
The insurance data is obtained from the Wharton School of Economics is WRDS service with the CRSP dailystock price database and contains daily pricing and return information from February 1990 to June 2020 ofinsurances with SIC codes from 6300-6499. The return data is then used along with the market capitalizationof each firm to create a market-cap weighted index of returns for the insurances. This insurance index willserve as a measurement of the entire insurance sector. The data is truncated for the same date range fromJanuary 2006 to June 2020 as mentioned in 4.2. Similar to all other data in this study, the daily returnsfrom the insurance index can be used to create return data for any length time period within the time range.This insurance data is important as it provides the insurance sector component of the financial industry thatwhen combined with other components from this study (hedge funds, banks, and brokerages) can be testedfor systemic risk within the entire economy using PCA (statistical test explained in 3.2).12
Results
In this section, the 2 systemic risk measurement methods discussed in Section 3 are applied to the S&P 500and Hedge Fund data mentioned in Section 4. Section 5.1 displays the results of cross-correlation calculationsconducted on the assets is return data, and Section 5.2 gives the results of PCA conducted on that data aswell.
The principal component analysis (hereafter referred to as PCA) is conducted between Hedge Funds, Broker-ages, Banks, and Insurances (all key financial asset types in the market). The 1st fractional eigenvalue is theideal data point to record as it is the most accurate measure of the concentration of variation between assetsis movements as the 1st fractional eigenvalue is too statistically unreliable and the fractional eigenvalues gointo the noise approaching 1 at the 3rd. Being the most accurate measure for concentration of variation inasset movements it is then also the ideal systemic risk indicator as increased concentration in the variationof assets means increasingly similar movements and market friction. With the 4 components of data beingconsidered as independently moving assets, the expected PCA fractional eigenvalues is an equal spread ofvariation among all assets/ components (fractional eigenvalues from 1-4 would be: .25, .5, .75, 1). However,this statistically expected data is highly unlikely, and any abnormal similarities in movements should createan increase in the value of the first component, indicating systemic risk across the entire financial marketas the assets being compared do span the entire financial industry. The PCA calculations and plot use thereturn data for the main LAB hedge fund index, as well as market-cap weighted indices of brokerages, banks,and insurances (for further explanation on the indices refer to Section 3.1) over the same 174-month timeperiod from January 2006 to June 2020. Figure 3 is the plot for the second fractional eigenvalues for thePCA on the 4 asset indices mentioned above, using 2 week return periods and 4 month (16 week) windowsfor each calculation over time.The plot from Figure 3 shows that from January 2006 to mid 2008 (periods 0 to mid 30’s), the firstfractional eigenvalue remains between .65 and .8. Although for complete statistical independence the firstfractional eigenvalue of 4 assets compared should be .25 it is understandable to have that high value above.5 when comparing movements of different financial industries as all industries should be moving in thesame general direction as the entire economy to some extent. Also considering that during that 2.5 yeartime period there was no market collapse, the .65 to .8 range recorded in that time period helps to set anexpected value of what the first eigenvalue should be in a normal market situation. Looking at the crashtimes from mid 2008 to 2012 (time periods mid 30’s to mid 70’s) shows a noticeable spike to values above13igure 1: Plot of first fractional eigenvalues of return data from indices of Banks, Brokerages, Insurers, andthe LAB hedge fund index (refer to sections 4.1, 4.3, 4.4, and 4.5 for further detail on data used) over rollingwindows of 30 time periods from over a 174 month range from January 2006 to June 2020..9 and even above .95, which are significantly outside of the normal range of .65 to .8. This spike only beginsto trend downward to the normal range during the periods following 2012 (time periods mid-70’s on onward)suggesting that the significantly high first fraction eigenvalues within the market crash time are related to thecrash and not a cause of an overall market change. Additionally, a high first fractional eigenvalue suggeststhat a lot of the variation in asset movements can be measured within the first principal component, showingthat the asset movements are very highly correlated. This significantly high correlation is another indicatorof high market friction (See section 3.2 for reasoning) between the 4 assets compared during the marketcrash period. Like in the results of the cross-correlations (section 5.1) this spike in market friction can berepresentative of a spike in systemic risk as well as the correlations between all assets would mean that eachasset can pose a threat to all others, making the group of industries very fragile and vulnerable to a systemicfailure which was observed during the time period. Mainly, the PCA results recorded in this test show aspike in fractional eigenvalue during the market crash from 2008-2012, which can be indicative of a systemicrisk within the network of assets tested for (hedge funds, banks, brokerages, and insurers) that needs to beeliminated in the future. 14 .2 Cross-correlations
This calculation of cross-correlations of Hedge Funds and the S&P 500 US market measure, display thecross-correlation coefficient between 2 of the given assets with a lag of one day to account for possible delayin systemically related movements. This cross-correlation method is conducted on the main LAB index aswell as its 5 sub-indices against the S&P 500 over a 174-month period from January 2006 to June 2020.Figure 1 provides the plots of cross-correlation coefficients over that time period between the LAB index andS&P 500.Figure 2: Cross-correlation coefficients plot (represented by blue line) with significance bands (red line)between the LAB index and S&P 500 over the 174-month time frame from January 2006 to June 2020.As seen in Figure 1, the cross-correlation levels between the LAB index and S&P 500 show movementsgenerally being part of the statistical noise within the significance bands of 21% (Calculated with lags = l = 90, significance bands = ± √ l ). The outliers of this trend are most frequent in the early 2007 to 2008time period, the same time as the market crash. During this period, the cross-correlation between the twoassets reached high 20% to low 30%, which is statistically significant. Soon after in mid 2008, it loweredfrom this significantly high cross correlation to a statistically significant negative cross-correlation of -28%for a short period of time. Regardless of the direction of the cross-correlation coefficients between the twoassets over that time period, the plot having frequent statistically significant cross-correlation between thetwo assets over that time period, may show that market friction is present during that time period alongwith significant, possibly dangerous levels of systemic risk. Thus, it is seen that systemic risk may haveexisted between the overall Hedge Fund Industry and the US market (represented by the S&P 500) duringthe 2007-2008 market crash. Considering these results, we can then plot the cross-correlation coefficients15etween the 5 subindices which may help to pinpoint the specific hedge fund sector that is responsible forthe possible systemic risk posed by the entire hedge fund industry on the US market. Figure 2 shows thecross-correlation plots between the 5 subindices and the S&P 500 over the same 174-month time periodplotted in Figure 1.Figure 3: Cross-correlation coefficients plot (represented by blue line) with significance bands (red line)between the GSL, EDL, LSL, MAL, and MFL hedge fund sub-indices and the S&P 500 over the 174-monthtime frame from January 2006 to June 2020.From the 5 plots, it is noticeable that not all sub-indices have similar statistically significant rangesof data similar to the overall LAB index comparison. Therefore, the data represented by the LAB indexcomparison must mostly be made up of statistically significant data from only the indices that share similartrends. Looking at that statistically significant time period for the LAB index from early 2007 to 2008,that similar frequently positive data outside of the significance bands ( 21% for all comparisons) is presentmostly in the GSL (high 20% to low 30%), EDL (high 20% to low 40%), and a little of the LSL (mid tohigh 20%) sub-indices during that same time period. The other 2 indices, the MAL and MFL mostly staywell within the significance bands during that period. Given that the LAB is made up of these sub-indices,we can then see that the statistically high cross-correlation data between the overall hedge fund industryand US market from that 2007-2008 market crash period is mostly created by those 3 hedge fund sectors16GSL, EDL, and LSL). Similar to the analysis of Figure 1, it is also possible that this cross-correlation is anindicator of market friction between those specific hedge fund sectors and the US market (represented by theS&P 500). Additionally, systemic risk may then also be associated with those sectors. Overall, the resultsof the cross-correlation calculations between the return data of the LAB and the 5 sub-indices it consists of,with the S&P 500, seem to show that statistically significant systemic risk was present between the overallhedge fund industry and the US market during the market crash from early 2007-2008. Specifically, the dataplots of the LSL, EDL, GSL sub-indices/ hedge fund sectors were mostly responsible for the significant LABdata during that time period, making those sectors the root of that possible systemic risk.The GSL index contributes largely to the significant cross-correlations noticed in the LAB, but withthis index being a variety of investment types that don’t belong in the other categories, it is difficult topinpoint where/ what investment strategies the possible systemic risk is stemming from and would requirea further investigation of the hedge funds included in the sub-indice and their relations with the rest of theeconomy. Looking at the definition of the LSL index from Section 4.1, it is seen that this group of hedgefunds engages in both long and short positions of investing. While long positions are not very risky financialmovements, short positions often are as they can be done with extreme leveraging and complex unregulatedfinancial instruments. According to the study conducted by Ferguson and Laster (Date), hedge funds usinglong/short strategies can maximize possible returns with higher leveraging and the use of obscure newfinancial instruments. These instruments would often in some way be related to other parts of the economysuch as investing in insurance premiums and using money lent from banks. In some way, hedge funds wereactually absorbing risk from the financial markets in a possibly healthy way. However, these instrumentscombined with the lack of regulation by the SEC in the early 2000’s allowed hedge funds to take extremelyrisky and high leverage positions on these investments that involved other institutions as well. Thus, whenthese hedge funds eventually failed as a result of excessive loss from high risk or sometimes frauds, the variousparts of the financial market they were involved with failed as well from the loss. In other words, the LSLstrategy hedge funds that had been absorbing a lot of market risk had failed and with that the risk spilledover to other parts of the financial market causing a systemic failure. Similarly, in the Event Driven LiquidityIndex (EDL) it can also be seen that hedge funds in these sectors also can develop potentially dangerousrelations with other institutions leading to increased systemic risk. As seen in the definition in Section 4.1,the EDL index consists of hedge funds using investment strategies based on the economic movements of shortterm events such as elections, company mergers, or bankruptcies. In the article written by Teun Johnston(2007) it is shown that these hedge funds have ”a dependence on external factors and in the short term, therecan be equity market relation”, showing that these some of the investment strategies of EDL index memberscreate market friction between themselves and other financial institutions, contributing to systemic risk.17dditionally, Johnston also explains that ”one of the major risks inherent in event-driven strategies is thatmanagers’ portfolios can be concentrated. Strategies such as merger arbitrage or corporate restructuring arehighly correlated to corporate activity and economic cycle”, showing how hedge fund involvement in suchevents creates major interactions between parts of the economy. This possibility of high market frictionalong with the potentially dangerous concentration in the portfolios of these hedge funds pose a threat ofsystemic failure as the crashing of these funds can easily spill over to the institutions they are involved with. With the advancements our society has made and is making, it is inevitable that our financial systems willbecome increasingly complex as well. The network between different financial institutions will become moreinterconnected as technology progresses. These increased connections as well as changes to the regulationsof our economy will provide new avenues for systemic risk to develop and pose a threat of systemic failure.Thus, it is crucial that we study this systemic risk and develop a map of its roots and causes to pinpoint theareas of our financial markets that need to have possible systemic prevention measures employed.The principal component analysis in this paper helps to measure and identify systemic risk throughanalyzing return data for evidence of market friction across Hedge Funds, Banks, Brokerages, and Insurers.From these tests we were able to conclude that spikes in market friction did occur between these financialinstitutions suggesting that systemic risk within the entirety of the US financial market was present duringthis time period and may have been a cause of the total failure. With identifying systemic risk within theUS financial markets to exist during crash time, we used cross-correlations to develop a map of possibleroots based on hedge funds, given that many other researchers had found hedge funds to be a likely stemof systemic risk. Our first cross-correlation test of the LAB Hedge Fund Index with the S&P 500 showedstatistically significant correlation to occur around crash time, meaning market friction and systemic riskwas present between Hedge Funds and the US Economy. We then further defined our map of hedge fundsby comparing each of the LAB is 5 sub-indices (EDL, LSL, GSL, MAL, and MFL) to figure out which ofthose indices were contributing most to the systemic risk found within the entire hedge fund industry. Ourresults displayed that the GSL, EDL, and LSL sub-indices were the hedge fund sectors mainly accountablefor the rise in systemic risk throughout the hedge fund industry which also spread through the entire USfinancial market.The results from the tests conducted highlight the need to evaluate the network and connections HedgeFunds have with other hedge funds and the rest of the market, especially the funds part of the GSL,EDL, and LSL indices. Eichengreen and Mathieson (1998) study these connections in relation to the 199718arket crash and Asian currency crash, developing a map of connections between hedge funds and theentire financial market. Their work is relevant in discovering the exact possible areas for the spread ofsystemic risk, helping to identify where stricter regulation may need to be. On the lines of regulation,having concluded from this study that areas of systemic risk do exist within the hedge funds sector, it is alsoimportant to study the regulations in place in these potentially dangerous sectors. Kambhu, Schuermann,and Stiroh (2007) conducted research on this topic amid the market crisis, analyzing the relations betweenCCRM (Counterparty credit risk management, see section 2) regulations and hedge funds to determinethe effectiveness of those measures and to which extent the failures of those measures created the systemicrisk and failure. From this study it is observed that changes to CCRM may need to be made in orderto prevent hedge funds and other financial institutions from taking potentially dangerous actions such asextreme leveraging and use of complex trading strategies that may not have been addressed.Our world will always keep advancing, making our financial networks more and more complex. Whilethese improvements are meant for the betterment of all, the complexity of them also creates the possibilitiesof more threats to our economy. It is crucial that a solid understanding of this increasingly complex networkbe understood and the necessary regulations be imposed on it to ensure the safety of our economy fromcomplete systemic failures. 19
References
1. Adrian, T., & Brunnermeier, M. K. (2009). ”CoVaR”. SSRN Electronic Journal. doi:10.2139/ssrn.12694462. Allen, F. (2001). ”Do Financial Institutions Matter?”
The Journal of Finance , 56(4), 1165-1175.doi:10.1111/0022-1082.003613. Billio, M., Getmansky, M., Lo, A., & Pelizzon, L. (2010). ”Econometric Measures of Systemic Risk inthe Finance and Insurance Sectors”. doi:10.3386/w162234. Chan, N., Getmansky, M., Haas, S., & Lo, A. (2005). ”Systemic Risk and Hedge Funds”. doi:10.3386/w112005. Chiang, T. C., & Li, H. (2007). ”Dynamic correlation analysis of financial contagion: Evidence fromAsian markets”.
Journal of International Money and Finance , 26(7), 1206-1228. doi:10.1016/j.jimonfin.2007.06.0056. De Bandt, Olivier, and Philipp Hartmann, 2000, “Systemic Risk: A Survey,” European Central BankWorking Paper No. 357. Eichengreen, B. J., Mathieson, D. J., & Chadha, B. (1998). ”Hedge funds and financial marketdynamics”. Washington, DC, District of Columbia: International Monetary Fund.8. Ferguson, R., & Laster, D. (2007).
Hedge Funds and Systemic Risk (p. 45-51). Banque De FranceFinancial Stability Review - Special Issue on Hedge Funds.9. Getmansky, M., Lo, A. W., & Makarov, I. (2004). ”An econometric model of serial correlation and illiq-uidity in hedge fund returns”.
Journal of Financial Economics , 74(3), 529-609. doi:10.1016/j.jfineco.2004.04.00110. Johnston, T. (2007). Event-Driven Strategies.
The Hedge Fund Journal , (27).11. Kambhu, J., Stiroh, K. J., & Schuermann, T. (2007). ”Hedge Funds, Financial Intermediation, andSystemic Risk”. SSRN Electronic Journal. doi:10.2139/ssrn.99590712. Kritzman, M., Li, Y., Page, S., & Rigobon, R. (2011). ”Principal Components as a Measure of SystemicRisk”.
The Journal of Portfolio Management , 37(4), 112-126. doi:10.3905/jpm.2011.37.4.11213. Lo, A. W. (2008). ”Hedge Funds, Systemic Risk, and the Financial Crisis of 2007-2008: WrittenTestimony for the House Oversight Committee Hearing on Hedge Funds”. SSRN Electronic Journal.doi:10.2139/ssrn.1301217 204. Muirhead, R. (1982). ”Aspects of Multivariate Statistical Theory”. Wiley Series in Probability andStatistics. doi:10.1002/978047031655915. Rickards, J. (2012, August 7). Repeal of Glass-Steagall caused the Financial Crisis.