Using Machine Learning To Predict Market Crashes
Small-cap, Mid-cap, Long/Short Equity, Growth
Seeking Alpha Analyst Since 2020
Leveraging alternative earnings-based valuation methodologies. Long-only ideas. Posts are the opinion of the author and are not investment recommendations. See 'Blog Posts' to apply these models to other securities.
Summary
- The Weak Form Efficient Market Hypothesis articulates that only new information changes the price of a stock.
- Stock market bubbles are the product of an inflated market value and when this corrects to fair value the market crashes.
- Stock market bubbles pop due to rapid changes in expectations caused by the inflow of new information.
- Machine learning can be used to accurately quantify the impact of new information on the stock market.
Introduction
If we can expedite the accurate analysis of new market information, we will be able to see bubbles before they pop. To understand the actual impact of new information on macro trends, we need to delve into the efficient market hypothesis and understand its relationship with stock market crashes.
The Efficient Market Hypothesis (EMH) states that stocks trade at their fair value, making it impossible to outperform the market. By the same token, inflationary bubbles cannot exist. Bubbles occur when the market value of a stock or industry diverges from its fair value. Under the EMH, this cannot happen. Unfortunately, history has indicated otherwise, but we can still find value from one form of the EMH.
Forms of the Efficient Market Hypothesis
- Strong Form - All information is ‘priced in’, including public and private information. Thus, all stocks trade at fair value regardless of private information. This is disproved by insider trading.
- Semi-strong Form - All public information is ‘priced in’, including past and present information. This nullifies fundamental analysis as a stock cannot be undervalued. The success of Warren Buffett and the majority of Portfolio Managers refute the validity of this form.
- Weak Form - Only past information is ‘priced in’, meaning technical analysis is pointless as the only thing that impacts a stock's price is new information. The Weak Form EMH is proved by the exitance of bubbles/crashes. A bubble occurs when a stock is incorrectly valued, and the bubble pops when there is "a rapid change in expectations based on new market information", causing the market value to self-correct. This rapid change is a reaction to current/new news, which (under this form) is the only thing that can alter a stock’s value, as past information is already priced in.
What Causes Market Value to Diverge from Fair Value?
We know from the past that the market is not completely efficient as stocks are often under/over valued. This value discrepancy comes from the misinterpretation of how new information will impact price. The amount of time an investor spends keeping up with financial/political news articulates that new information changes the price of a stock. It is investors’ inability to accurately quantify the impact of current news that creates differences in the perceived value of a stock.
The Weak Form Efficient Market Hypothesis assumes investors have access to all relevant information and that investors make informed investment decisions. The entire concept of Behavioral Finance refutes this as investors often fall victim to bias (overconfidence, herd mentality, loss adverse). People interpret news differently, making valuation largely subjective. We can hedge against this bias by removing interpretation from news quantification. A standardized model to analyze financial/political information could be used to not only find the correct change in a stock’s fair value but to also predict the market, as automation would replace interpretation.
Machine Learning to Quantify Valuation
Currently, automated trading is primarily based on technical analysis, which is not valid under these assumptions. We are only using machine learning to execute actions based on preexisting parameters; automation should be used to quantify financial/political news. This would remove interpretation and guesswork from predicting the market and would calculate the change in value before/as it is applied to a stock’s price.
This is currently being tested on a macro level by the Reserve Bank of Australia. They created the News Sentiment Index, which analyzes news using machine learning. They scan the language of news articles for positive/neutral/negative words and quantify the outlook of the entire article. In theory, news analysis should be able to accurately predict changes in the stock market; this prototype was successful enough to correlate with macro trends and preemptively align with economic indicators.
RBA’s News Sentiment Index
In their research report, RBA writes “Not only is the NSI a very timely indicator, but changes in news-based sentiment precede movements in survey-based sentiment measures. This may be because, in making decisions, consumers and business managers rely on high-frequency information that is broadcast through the news media.” This means that by pulling the sentiment of new information instead of leaving it up for interpretation, their index correctly quantified the impact of news changes before investors could.
The following chart plots their NSI against the market (NSI in dark purple). We can see that new information is what changes the market; currently, we just fail to interpret this information quickly and accurately. The NSI shows a strong correlation with the market and is successful on a macro level.
Source: Reserve Bank of Australia's News Sentiment Index Research Report
The following two charts evaluate to what extent the NSI was able to predict bubbles/crashes. Since the pop of a bubble is characterized by "a rapid change in expectations based on new market information", this is where quantifying the news would have the most utility. The NSI (orange) acted as a strong predictor of the COVID 19 downturn, which began in March 2020 (lower chart). The NSI also reflected the beginning of the recovery from the 2008 financial crisis before business surveys articulated the same optimism (upper chart).
Source: Reserve Bank of Australia's News Sentiment Index Research Report
Conclusion
Overall, the Weak Form Efficient Market Hypothesis is proven by the existence of bubbles/crashes and tells us that only new information impacts the market. From this, we understand that bubbles pop when this news information has a large impact on expectations. This change in expectations is the catalyst for market value to self-correct, leading to a crash. Now that we have a better understanding of what pops market bubbles, we can learn how to predict them. Through machine learning, we can remove interpretation from the analysis of financial/political news and see its actual impact. Thus, we can use automation to find value discrepancies and lower our propensity to overreact to changes in the market.
Analyst's Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.
Seeking Alpha's Disclosure: Past performance is no guarantee of future results. No recommendation or advice is being given as to whether any investment is suitable for a particular investor. Any views or opinions expressed above may not reflect those of Seeking Alpha as a whole. Seeking Alpha is not a licensed securities dealer, broker or US investment adviser or investment bank. Our analysts are third party authors that include both professional investors and individual investors who may not be licensed or certified by any institute or regulatory body.