What decisive factors have been driving the stock markets since 2017? IT IS NOT Economic Indicators

Abstract

Stock prices fluctuate according to changes in supply and demand. Investopedia, known as an extensive collection of online financial resources, listed the most common catalysts driving daily stock price movements: fundamentals, technical factors, and market sentiment. This study will be analyzing which factors mostly correlate with fluctuations in stock prices. The independent variables this study will use are EPS and PE ratios, which will be representing the fundamental aspects of stock prices; technical factors, represented by CPI and GDP numbers as well as substitute asset classes demand (government debt and median price of houses sold), and finally the VIX index which will monitor sentiment. The results of the study will serve the purpose of not only helping people better understand market fluctuations and which factors have a higher influence on stock prices but also create a model that can predict how prices will behave under certain market conditions. Anyway, this study considers the S&P 500 daily share price as the benchmark to representing the dependent variable (stock prices) simply because the index is considered the best representation of America’s stock market, with the country’s 500 biggest companies by market cap being quoted on the exchange. In essence, the study results show that although some results proved to be as expected, such as stock prices being positively correlated to positive economic news, many results proved to reject initial hypotheses, such as the VIX and alternative asset demand being positively correlated to stock prices or fundamental factors having a negative correlation. Lastly, economic news and fundamentals proved to be less influential towards stock prices than what was previously believed.

Introduction

            Although many people consider the stock market a way to easy riches, being profitable in the market is an enigma very few seasoned investors have managed to solve. According to a study led by eToro, a large European multinational broker, 80% of their 27 million active users lost money in the stock market, averaging -36.3% returns over 12 months (Lyck, 2020). Therefore, the stock market is all but easy to trade, reasons ranging from stock prices behaving in a random fashion, investors misinterpreting information, or the efficient market theory keeping traders from making money long-term. Whichever the reason, this study strives to understand which factors have the largest impact on S&P 500 prices, so that investors can improve their chances of better predicting price movements according to the different fundamental, technical and sentiment market environments.

            The next section will have a literature review that will analyze any past research led on the topic of stock price fluctuations and their catalysts. Knowing additional information on the topic may make it easier to form hypotheses about the results of this study. Secondly, there will be a section that discusses the study regression model, including any information related to data and methodology. Thirdly, will come a section that discusses the regression results and analyze any special characteristics that could have tweaked the results. Lastly, we will interpret the final data and understand what conclusion can be made about stock market price variations from these results.

Literature Review

What drives stock price movements? Review of Financial Studies

            A journal written by Chen et al. in 2013, addresses the most important factor that the public believes move stock market prices: expected earnings. In fact, the article does this by analyzing how sensitive stock prices are to expected company cash flows by computing numbers using the implied cost of equity capital (ICC) method at a firm and aggregate (market) level to generalize results to the wider stock market indices. One upside of this study is that the ICC method the researchers employ during their study is a superior model to the predictive regression method which they discuss about in their journal but refuse to use due to the inefficiencies it may produce to the study results. For example, the predictive regression method is sensitive to sample period, predictive variables, and cash flow inputs. Despite that, a clear inefficiency of the journal is the independent variables it includes, which may not be directly comparable to the variables used in the research of this paper. For example, Chen et al. use expected cash flows as the explanatory variable, while this study uses EPS, and PE ratios. Another issue that might impact results is the fact that the study employs cash flow forecasts rather than hard data to lead research, making the study unpredictable to a certain extent.  

The results of the study show that stock returns are greatly impacted by the cash flow news component with the greatest impact on price for investment horizons over two years.

What moves stock prices?

            A second example of a paper written by David et al. that tries to estimate the variance in stock prices according to certain kinds of news concludes that stock prices are greatly affected by a diverse variety of catalysts. Firstly, non-public information greatly impacts stock prices as larger price moves mainly occur during days without any identifiable major news release, casting doubt that predictable stock catalysts mentioned above, such as future cash flows are as effective an indicator for price moves. Thirdly, it is shown that non-economic news has no significant impact on stock prices, therefore showing that the main news drivers of stock prices are economic data releases. The study yields the abovementioned results by physically analyzing how major news impacted S&P prices in the period ranging from 1941 and 1987, but also by using models to understand how much of a variation in stock prices is explained by news, mainly through using the coefficient of variation, and VAR evidence.

            In essence, this paper by David et al. relates to my study as it gives an insight into a different catalyst that potentially impacts stock prices, news type. Additionally, by showing that economic data is the leading news driver to stock prices implies that the dependent variables, such as CPI and GDP numbers employed in my study will most likely be positively related and be significant drivers to changes in dependent variable (S&P500 prices).

Data and Methodology

            This study stives to analyze the best regression model to estimate the correlation between the explanatory variables mentioned in the abstract and the dependent variable (S&P500 price). After having designed the most appropriate regression model for the price, this study strives to understand which of the explanatory variables causes the largest change in y (is the leading catalyst of stock price movement) with the goal of allowing investors to better understand which factors drive stock price movements and to better predict stock price behavior under certain macroeconomic conditions. As shown by the sources quoted in the literature review, stock prices are highly correlated to future cash flow numbers and economic news, while being, virtually, unaffected by non-economic news. After having compiled the seven most accepted catalysts to stock price changes and having analyzed past research on the topic, I expect nominal GDP, EPS, PE ratios, and CPI numbers to have the largest positive correlation to SP500 price. The study uses OLS methodology (ordinary least squares) to yield the best predictive model through minimizing the sum of squared residuals and employing a 5% significant level to decide if data in the prediction model is statistically significant.

            Figure 1 shows the daily S&P 500 price since November 2017. The data is retrieved from finance.yahoo.com and is measured in US dollars per share of S&P500. As shown in figure 1, the S&P price frequency is clustered mostly between the 2000-3000 mark. However, figure 2 shows how in the years the price of the index entered a rally that carried its price from a low of 2237.4 to a high of 4796.56.

            To understand price change, various datasets were gathered from time series data sources for the period 2017-2022. Stocks are moved by three possible catalysts fundamentals, technical factors, and market sentiment signs. Major fundamental indicators include earnings per share (EPS) and price-to-earnings ratio (PE). Data for both EPS and PE ratios are taken from Ycharts.com, a source that specializes in time series financial data, and are shown monthly since 2017. Both financial ratios are generated from only S&P500 companies, so that overall statistics of the index can better emulate original data. Figures 3 and 4 show the distribution histogram for the S&P500 PE ratio and EPS ratio accordingly. Like Figure 1, both histograms are right-skewed, meaning that at a first glance, S&P500 prices might be largely linked to these financial ratios. 

            Technical indicators include CPI numbers, nominal GDP numbers, as well as alternative asset demand, such as public debt and real estate prices. The CPI data was retrieved from the U.S. Bureau of Labor Statistics and is measured on a percent basis monthly since 2017. Data for nominal GDP was taken from the Federal Reserve Economic Data website and is measured in billions of dollars on a quarterly basis that dates to January 2017. Past research shows that economic news has a great impact on stock prices, so I expect GDP data and CPI numbers to have a large impact on stock prices. Data for alternative asset demand is retrieved from Savills.com which tries to estimate the market capitalization of different assets. As shown by the source, total equities value stands at $109.2 trillion with both real estate and the debt obligations market dwarfing its size with a value of $258 trillion and $123.5 trillion accordingly. Therefore, it is important to understand the relationship between equities demand and alternative asset demand to see if a change in demand for other assets affects demand for equities and, therefore, changes stock prices. Data for real estate prices is retrieved from Ycharts.com and the US existing home median sales price in USD since 2017 is used to measure the variable. While to understand the variable of debt obligations, the data is retrieved from the Federal Reserve Economic Data Website that supplies the information for daily nominal outstanding debt (notes and bonds) on a weekly average measured in millions of dollars. The idea with alternative asset demand is that with an increasing alternative asset demand, less money is available to invest in equities, which drag S&P500 prices lower. Therefore, the correlation will be negative.

            The VIX is an index that represents S&P500 sentiment through expectations of the relative strength of short-term price changes, which also allows traders to make a 30-day projection of future volatility. The VIX is also called the fear index because volatility increases as investors start doubting market conditions, therefore, making the VIX a valuable sentiment indicator. VIX information is retrieved from Cboe.com and is measured monthly in points which are calculated by combining the weighted prices of index put and call options for the index in a period of 30 days. The idea is that the VIX will be negatively correlated with the S&P500 because when investors are fearful, VIX prices tend to increase while S&P500 prices tend to be suppressed by lower demand.  

Summary statistics for all explanatory variables are shown in figure 5.

Regression Results

            Although the data and methodology section stated that the regression model for this paper would include seven explanatory variables, there were complications upon testing that forced the model to change to accommodate and improve the results. The final regression model is:

*

*The regression model accounts for the exclusion of influential values found in the study

Analysis of assumptions of the error term

A series of tests were run to test for the violation of the four assumptions of the error term used in the regression model.

Linearity Assumption

            Figure 6 shows the graph of residuals of the regression to fitted values, a scatterplot that shows no correlation between the points. This means the explanatory variables have a linear relationship.

Homoskedasticity

            The test for homoskedasticity includes a scatterplot of residuals that shows no cone-shaped formation and a Breusch-Pagan test that detects no heteroskedasticity. Figure 8 shows the test result.

Normality Assumption

            Figure 9 shows the results of the Shapiro-Wilk test for normality. Unfortunately, the result allows us to reject the null hypothesis of normality, which does not allow us to confirm normality for the model. Although normality cannot be proved through test, normality can be proven through the Central Limit Theorem. For our model, we should have about 70 observations to prove normality through the theorem. Therefore, normality is proven through the Central Limit Theorem.

Independent Error Terms
            The Durbin Watson Test yields a result that rejects the null hypothesis of no autocorrelation. A possible solution to this is to use either first differing or a Prais-Winsten transformation. Nevertheless, both transformed results yield indeterminate results, meaning that we cannot assume that the error terms are independent.  

Influential Observations

            The regression model also included three influential observations, all of which were signaled a Cook’s D statistic test as shown in figure 9. The three most influential values were also justified in figure 7, where 3 points are marked as outliers outside the -2 to 2 standard deviations in the graph of standardized residuals. These influential observations were removed from the regression model.

Multicollinearity

            At first, multicollinearity was a problem in the regression model that included all seven explanatory variables initially mentioned in the paper. However, after careful examination of the regression model and variables, GDP, EPS, and Realestateprice variables were responsible for the most collinearity in the model. These variables could be omitted from the final model to fix multicollinearity, while also being replaced by other variables that emulated the former’s implications. For example, EPS numbers are like PE numbers and the relationship the regression model analyzes in the latter, can be expected in EPS alike, both positive. In other words, EPS can be expected to behave similarly to PE ratios in the regression model. Other examples include the demand for an alternative asset class like real estate where demand can easily be measured from bonds or the economic news characteristic of GDP where good economic data can pretty much be observed in CPI numbers and where the use of both is unnecessary, especially considering that GDP was one of the explanatory variables that skewed multicollinearity the most. In essence, many of the explanatory variables in the original model could be described as superfluous as they likely reiterated results another closely related variable already included in the model would. Final multicollinearity numbers were tested through Stata and the results can be seen in figure 10. No variable has multicollinearity over 5.

Results

            First, figure 12 shows that all variables, except for CPI, are statistically significant at a 5% confidence level. However, by excluding the CPI variable from the model, r-squared would be negatively impacted, so the variable will not be removed. Figure 11 regression (4) shows the full model for this paper. The regression results suggest that the model explains about 60% of the variation in S&P500 price from the explanatory variables. As can be seen from the figure, adding the other explanatory variables that were removed later in the study make the results extremely accurate with an r-squared of 0.895 in regression (1). However, the problem would arise with proving most OLS assumptions, yielding a model that is accurate on the surface but impractical. In the full regression model, the only explanatory variable that is positively correlated with the logarithm of S&P500 prices is the amount of bonds, such that for every million dollars increase in the amount of bonds outstanding, the logarithm of the price of each S&P500 share increases by 0.00000148 percentage points. Despite being negatively correlated, regression (4) shows that the explanatory variables most strictly related to the logarithm of S&P500 prices are VIX, PE ratios, and CPI, in the sequence in which they are listed. For every dollar increase in the logarithm of SP500 prices, there was a decrease of 0.00245 percentage points in VIX.

            Regression (3) eliminates the logarithm for the dependent variable, showing how explanatory variables are correlated to S&P prices. Firstly, it is unexpected to see PE ratios and EPS negatively correlated to the S&P. Expectations were that as S&P prices increased, so would current and future earnings in an optimistic investing environment. A possible reason for this behavior could be that with higher prices, PE ratios are getting dragged down by decreasing expected earnings per share, which are affected either by an environment which greatly overestimates companies’ net income during overoptimistic stock market runs, where companies are, additionally employing more of their capital and earnings to expand activities and start new ventures or where earnings per share are decreased by companies issuing more stock during periods of high prices (market rallies). Secondly, economic news seems to have been correlated to a smaller extent compared to what was predicted, while the sentiment indicator VIX was the most correlated. The reason for this behavior could be that economic data does not reflect the complete picture for investor sentiment, after all different investors might interpret economic data differently, while the VIX truly shows how investors feel towards the current market.

Conclusion

            This study sought to find which variable, among seven that investors consider the most impactful on stock prices, were the most correlated to the S&P500. The results of this study could be used by investors to not only understand what macroeconomic conditions move jointly with stock prices, but also better understand market conditions and what can be expected of stock prices under different economic environments.  The explanatory variables were nominal GDP numbers, CPI numbers, average S&P company EPS, average S&P company PE, VIX prices, median house sales price, and outstanding public debt. Results concluded that the sentiment indicator, VIX, had the largest correlation to stock price movements, followed by fundamental factors, such as PE ratios, economic news, such as CPI, and alternative asset demand represented by bonds. Past research was used to predict that economic news and expected future earnings had a positive correlation to stock prices. Nonetheless, the results showed that the relationship between these factors was weaker than expected and sentiment indicators, such as VIX was better correlated. One finding that also did not align with one of the initial hypotheses was the impact of alternative asset demand. Initially, alternative asset demand was supposed to have a large correlation to stock prices as investors must choose between either buying stocks or investing in alternative assets, such as real estate or public debt. The results of the regression proved otherwise. Although correlation between alternative asset demand and stock prices was weak, the correlation was positive nonetheless, meaning that stock prices increased as investors demanded more bonds and real estate investments. A reason for this might be that either investors who invested in other classes, also subsequently invested in stocks to diversify, hereby driving up stock prices or people tended to invest in alternative asset classes when the investing environment was ideal at the same time that stock prices were increasing.

            Finally, the research also showed many shortcomings. Firstly, research quoted in the literature review that was outdated, articles written in 1988 and 2013, which analyzed stock prices in times, where a different environment might have impacted stock prices. In other words, the research in this paper was led with data from the period 2017-2022, which might not align the research led in years 1988 and 2013, meaning the predictions made with outdated articles might be inaccurate. Secondly, the economic news used in the model, GDP, and CPI, are very limited in showing their impact on stock prices for many reasons. First, although GDP and CPI are considered some of the most important economic indicators of the stock market, different market conditions may weigh other economic indicators more heavily at different times. Second, other non-public economic or government manipulated data might be affecting stock prices, which might be affecting stock prices without letting investors know. Third, the regression model posed a challenge, whether to eliminate some explanatory variables which posed a problem with the properties of ordinary least squares or whether to leave these variables to significantly increase the r squared.

References

Assets: Securities held outright: U.S. Treasury Securities: Notes and bonds, nominal: Week average. FRED. (2022, December 1). Retrieved December 6, 2022, from https://fred.stlouisfed.org/series/WSHONBNA

Chen, L., Da, Z., & Zhao, X. (2013). What drives stock price movements? Review of Financial Studies, 26(4), 841–876. https://doi.org/10.1093/rfs/hht005

Cutler, D. M., Poterba, J. M., & Summers, L. H. (1988). WHAT MOVES STOCK PRICES? . NBER WORKING PAPER SERIES .

Gross domestic product. FRED. (2022, November 30). Retrieved December 6, 2022, from https://fred.stlouisfed.org/series/GDP

Harper, D. R. (2022, July 28). What drives the stock market? Investopedia. Retrieved December 4, 2022, from https://www.investopedia.com/articles/basics/04/100804.asp

Historical Data for Cboe VIX® Index and Other Volatility Indices . Cboe Global Markets. (2022). Retrieved December 6, 2022, from https://www.cboe.com/tradable_products/vix/vix_historical_data/

Larson, S. J., & Madura, J. (2003). What drives stock price behavior following extreme one-day returns. Journal of Financial Research, 26(1), 113–127. https://doi.org/10.1111/1475-6803.00048

Lyck, M. (2020, December 28). Why 80% of day traders lose money. Medium. Retrieved December 4, 2022, from https://marklyck.medium.com/why-80-of-day-traders-lose-money-78d51b10fe25#:~:text=According%20to%20the%20stock%20platform,quitting%20within%20just%20two%20years.

S&P 500 Earnings Per Share. S&P 500 earnings per share. (2022, July 1). Retrieved December 6, 2022, from https://ycharts.com/indicators/sp_500_eps

S&P 500 PE Ratio by Month. S&P 500 pe ratio by month. (2022). Retrieved December 6, 2022, from https://www.multpl.com/s-p-500-pe-ratio/table/by-month

Tostevin, P., & Oakley, M. (2021, December 7). The total value of Global Real Estate. Savills Impacts. Retrieved December 7, 2022, from https://www.savills.com/impacts/market-trends/the-total-value-of-global-real-estate.html

U.S. Bureau of Labor Statistics. (2022). Consumer price index historical tables for U.S. city average. U.S. Bureau of Labor Statistics. Retrieved December 6, 2022, from https://www.bls.gov/regions/mid-atlantic/data/consumerpriceindexhistorical_us_table.htm

US Existing Home Median Sales Price. US existing home median sales price. (2022). Retrieved December 6, 2022, from https://ycharts.com/indicators/us_existing_home_median_sales_price

Yahoo! (2022, December 6). S&P 500 (^GSPC) historical data. Yahoo! Finance. Retrieved December 6, 2022, from https://finance.yahoo.com/quote/%5EGSPC/history/?guccounter=1&guce_referrer=aHR0cHM6Ly93d3cuZ29vZ2xlLmNvbS8&guce_referrer_sig=AQAAALK1EOooWZOsuEhwGxMWtKYaErDH9u4V_waeemkJYEat5Pvpxl2jeE3CS1r5G6Uf79brojhyy5ArIk1S7V0ck98TPOhK28g-F1wRzRQOr0mcC-hribVCfvKKz5_yEHxxtGYf2ZtzvkTTuStEUrDFM63PEK7Ah835Ic4-ARKX2yE2

Appendix

Figure 1- Histogram of daily price of S&P500 index since November 2017

Figure 2- Summary statistics of S&P500 price in the period 2017-2022

Variable	Obs	Mean	Std. dev	Min	Max
Sp500price	1259	3424.475	674.2795	2237.4	4796.56

Figure 3- Average S&P500 PE ratio over the period 2017-2022

Figure 4- Average S&P500 EPS ratio over period 2017-2022

Figure 5- Summary statistics for regression model explanatory variables

Variable	n	Mean	S.D.	Min	.25	Mdn	.75	Max
Sp500price	1259	3424.47	674.28	2237.40	2822.48	3230.78	4023.61	4796.56
real estate price	50	3.2+05	49362.06	2.5+05	2.7+05	3.1e+05	3.6e+05	4.1e+05
VIX	2191	99.41	17.56	61.76	86.60	96.28	110.29	207.59
Gdp	23	21683.48	1919.17	19148.19	20155.49	21362.43	23046.93	25663.29
CPI	70	261.39	15.24	242.84	251.59	256.87	269.20	298.01
PE	72	24.54	4.56	19.38	21.65	23.20	25.16	39.26
EPS	50	27.94	9.15	11.88	21.72	24.87	33.02	53.94
Bonds	1041	1.7e+06	1.3e+06	3.9e+05	4.6e+05	1.6e+06	2.3e+06	5.0e+06

Figure 6- Residual to Fitted Values (linearity test)

Figure 7- Standardized Residuals to fitted values (Heteroskedasticity test)

Figure 8- Breusch-Pagan test for heteroskedasticity

Figure 9- D Cook’s statistic test

Figure 10- Multicollinearity Results for Regression Model

Variable	VIF	1/VIF
PE	1.12	0.895076
VIX	1,10	0.906977
CPI	1.07	0.932951
Bonds	1.06	0.946335
Mean VIF	1.09

 

Figure 11- Full regression model results

Variables	(1) sp500price	(2) sp500price	(3) sp500price	(4) sp500price
Vix	0.222	-0.206	0.772	-0.00245***
Cpi	0.1		0.268	-7.14e-05
Bonds	0.0116***	0.00820***	0.00964***	1.48e-06***
Eps	-0.359		-0.0141
Pe	-6.484*		-8.669**	-0.00130**
Gdp	0.00949**	0.00603	0.0109**
Realestateprice	0.000520*	0.000540**
Constant	-2,184*	-877.2	-1,290	7.537***
Observations	22	22	22	66
R-squared	0.895	0.856	0.861	0.358

Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

 

 

 

Figure 12- Regression results table