This report is a statistical analysis of the validity of using stocks with negative beta coefficients as hedges against economic downturns. The scope of this report does not include financial instruments which use short positions to artificially create inverse relationships.
Does an equity with a negative beta coefficient in a bull market truly make a worthwhile hedge in a bear market?
Conventional stock market wisdom suggests that stocks which have inverse relationships with overall market performance may have a place in an investor's portfolio if said investor believes the overall market is going to experience a moment of decline (Streissguth, 2019). Mathematically, this theory makes sense; however, it is unclear if the theory holds water in real-life applications. A majority of stocks with negative beta coefficients may simply be underperforming or non-performing assets that continue to perform poorly in a downturn.
The hypotheses and null hypothesis are as follows:
H0: Stocks with negative beta coefficients perform as expected during distressed market conditions (there is no statistically significant change in beta coefficients).
H1: Stocks with negative beta coefficients perform better than expected relative to market indices during distressed market conditions (beta coefficients decrease).
H2: Stocks with negative beta coefficients perform worse than expected relative to market indices during distressed market conditions (beta coefficients increase).
Data collection for this research paper began by finding samples to test the hypotheses. Ideally, one would be able to analyze complete data of historical prices for every equity that has ever traded publicly; however, this paper will be analyzing sample sets instead.
A representative sample began to be gathered by manually scraping mentions of stocks with negative betas from the web. To obtain a valid representation, examples were taken from different time periods, different industries, and different beta ranges. Various search terms were used when scraping the web to try and mitigate confirmation bias. There was a total of 55 observations recorded with 49 of them being distinct.
As some of the observations were American Depositary Receipts (ADRs), it also became necessary to retrieve historical currency exchange rates to isolate for interactions between the equity and the index rather than interactions between the U.S. Dollar and the native currency. This data was retrieved from ofx.com (“Historical Exchange Rates Tool & Forex History Data”, 2020).
The next step was to retrieve historical price data for each of these observations so that beta coefficients could be calculated. To do this efficiently, a node.js script was developed which used the Alpha Vantage API. The code is as follows:
const https = require('https');
Finally, the data for the S&P 500 (NYSE: SPY) was retrieved using the node.js script.
Data Extraction and Preparation
The first step in the data-preparation process was to put all historical price for each ticker into a single, readable table where the data can be compared easily. This step exposed fourteen observations which did not contain enough data to be used in the analysis (DPS, ENEVY, GRSXY, GEGYY, DRCSY, PYPTF, CTWS, CHCJY, KGDEY, TRMD, ZM, FRHC, AHCO, PLMR). This was a relatively easy process, so it was performed manually in Excel.
The resulting table begins as follows and continues through 08/31/2000:
The advantage of preparing this data in Excel is that it isn't necessary to spend time creating a script to prepare the data; however, if the amount of data were larger, it may make sense to spend the time writing a script rather than spending time manually manipulating the data. Setting the data up in a table like this will allow for each column to be read in as a variable in R for regression to be performed. Additionally, the regression can be performed directly in Excel. This format also sets the data up nicely for Excel-based regression to be executed.
The data for the independent variable (S&P 500) was then appended to the table:
For quick reference, ADRs were then color-coded by country. Each country was then given a corresponding Excel tab with a matching color which contains the historical exchange rates for the country's native currency. Setting the data up this way allows for easy access to desired information through tools like Excel's “VLOOKUP()” function.
These steps are illustrated by the following images:
While linear regression is how the beta statistic is commonly calculated within the industry, it can also be calculated simply by dividing covariance by variance (Nickolas, 2019). However, this analysis will be using complete linear regression models to dive more deeply into the strength of the relationships and evaluate more fully the null hypothesis that the difference between beta coefficients is actually zero. Additionally, linear regression is easier to use when evaluating models with multiple independent variables. Out all of the tools I've used throughout this program, I've found that R is the easiest to use when performing regression methods on larger data sets
This analysis will be focusing on two “inflection points” from which the market turned from a positive trend to a negative trend. Specifically, these two inflection points are October 31st, 2007 and December 31st, 2019. The next step was to calculate the Beta Coefficient (3Y monthly) from each of those points for every one of the dependent variables. This was done using linear regression in R with the following code:
The results are presented below:
As can be observed, the first inflection point narrowed our data down to only eight observations with negative beta coefficients. These eight stocks would likely move forward to the next step of analysis; however, two of the stocks were ADRs which meant that the currency exchange rate had to be factored into the regression model.
To accomplish this, the following code was executed:
ADR <- read_excel(“C:/Users/Frenc/OneDrive/Desktop/ADR.xlsx”)
ADR.xlsx combined the previously gathered exchange rate data with the relevant stock variables and was limited to only the desired timeframe. This code produced the following results:
As can be observed, FLRAF's beta changed from -0.00595 to -0.007924 and IHICY's changed from -0.02372 to -0.02170 after taking exchange rates into account.
Further review and verification revealed two more ADRs in the list: BUD and COE. As such, the following code was executed to adjust for currency interactions:
budRegression <- lm(ADR$BUD ~ ADR$SPY + ADR$EURO)
It is important to note that simply specifying the exchange rate in the model may not consider exogenous effects which occur when currencies strengthen and weaken against each other.
This left the analysis with the following eight observations:
The ADR dataset was then updated to include the post-inflection data and the flowing code was executed:
#DETERMINE POST-INFLECTION BETA
The final data for the first inflection point is as follows:
The same process was then performed for the second inflection point. After running the following code, only four data points were found to have a negative beta coefficient (3Y monthly) preceding the inflection point.
inflectionTwo <- output[154:190, ]
The four observations were DCMYY (-0.0038776), WMT (-0.03485316), AMGN (-0.1831979), and HRB (-0.03964931). DCMYY's regression model was adjusted to include currency interaction between the USD and JPY using the following code:
ADR2 <- read_excel("C:/Users/Frenc/OneDrive/Desktop/ADR2.xlsx")
This resulted in the following output:
As the beta coefficient turned positive when the exchange rate was included, DCMYY was excluded from further analysis. The next step for the other three observations was to gather the post-inflection data. This was accomplished using the code displayed below:
#GATHER BETA COEFFICIENTS FOR THE SECOND POST-INFLECTION POINT
The final data for the second inflection point is included in this table:
Data Summary and Implications
Testing the hypotheses of this analysis was done by performing a Paired sample T-test, using T distribution (DF=10) (two-tailed) with the following code:
before <- c(-0.05472,
The results are displayed below:
This demonstrates that the null hypothesis which states that there is no change in beta coefficients after an inflection point can be rejected at at least the 95% level. The mean difference between pre-inflection and post-inflection data was -0.07187598 meaning that, on average, beta coefficients increased by more than 0.07 after an inflection point. This shows preference for H2 which states that stocks with negative beta coefficients perform worse than expected relative to market indices during distressed market conditions (beta coefficients increase).
The limitations of this analysis include a relatively small sample size with possible bias in the selection process (a truly representative sample may not have been achieved). However, based on this analysis alone, it is recommended that investors seeking to use equities with negative coefficients as a hedge against overall market downturns expect their assets to become more correlated to the market in a downturn.
Two approaches which may be appropriate for future study would be to obtain an entire population to study rather than a sample and to separate observations by the reasons they might have an inverse relationship with the market such as the stock being counter-cyclical or under-performing. This would allow each regression model to be more correctly specified.
Streissguth, T. (2019, March 31). Is a Negative Beta Coefficient More Risky Than a Positive in the Stock Market? Retrieved June 27, 2020, from https://finance.zacks.com/negative-beta-coefficient-risky-positive-stock-market-7596.html
Nickolas, S. (2020, February 05). The Formula for Calculating Beta. Retrieved June 27, 2020, from https://www.investopedia.com/ask/answers/070615/what-formula-calculating-beta.asp
Historical Exchange Rates Tool & Forex History Data. (2020). Retrieved July 13, 2020, from https://www.ofx.com/en-us/forex-news/historical-exchange-rates/
Caplinger, D. (2012, December 12). Negative-Beta Stocks: Worth Buying? Retrieved July 14, 2020, from https://www.fool.com/how-to-invest/2012/12/12/negative-beta-stocks-worth-buying.aspx
Negative Beta Stocks. (2020). Retrieved July 10, 2020, from https://www.marketbeat.com/market-data/negative-beta-stocks/
Hacker, D. (2008, January 28). Can beta be negative? Retrieved July 10, 2020, from https://www.analystforum.com/t/can-beta-be-negative/5351
Carnevale, C. (2015, May 21). The Great Beta Hoax: Not an Accurate Measure of Risk After All. Retrieved July 10, 2020, from https://www.advisorperspectives.com/commentaries/2015/05/21/the-great-beta-hoax-not-an-accurate-measure-of-risk-after-all
Sanchez, CFA, L. (2020, January 15). This Countercyclical Stock Could Be a Good Hedge. Retrieved July 10, 2020, from https://www.fool.com/investing/2020/01/15/this-countercyclical-stock-could-be-a-good-hedge.aspx
Delaet, R. (2020, May 15). 3 Strong Hedges For A Second Market Crash. Retrieved July 10, 2020, from https://seekingalpha.com/article/4347970-3-strong-hedges-for-second-market-crash
Divine, J. (2020, March 18). 7 Stocks That Soar in a Recession. Retrieved July 10, 2020, from https://money.usnews.com/investing/slideshows/7-stocks-that-soar-in-a-recession?slide=2