Long/Short Equity Strategies And Machine Learning

Parts of this article were originally published in Price Action Lab Blog.

Long/short equity strategies are preferred by hedge funds since their performance does not depend on an explicit forecast of market direction.

The general idea when executing a long/short strategy is to hold N securities long and M securities short, usually with M = N, and realize a net gain on the trade. Although this type of strategy offers maximum potential when both long and short positions profit, it is also profitable when long positions are more profitable than short, or the other way around. When the long and short invested amounts are equal, this strategy is usually referred to as “market neutral” but this should not be interpreted as “risk-free.”

Long/short equity strategies minimize market exposure but may fail due to flaws common to all strategies, for example errors in analysis, over-fitting to past conditions, selection bias, data-snooping bias, just to name a few.

Most retail traders are not familiar with long/short equity strategies because they are primarily interested in forecasting price direction. Other than passive buying and holding, all trading strategies with outcomes that depend on the direction of prices make explicit or implicit forecasts although some traders, mainly trend-followers, may argue otherwise.

Retail traders often use a trivial version of long/short equity with N = M =1, known as a “pair trade.” They usually take a long position in a stock from a sector that outperforms the market and a short position in a stock from a sector that has negative performance, in equal dollar amounts. Risk is maximum when both legs of the trade turn unprofitable.

Most hedge funds that employ long/short equity use fundamental analysis to identify companies, sectors and even countries to go long or short.

In recent years, there is growing interest in using machine learning to develop and execute long/short equity strategies. For example, Point72, which invests the assets of Steven A. Cohen and eligible employees, “primarily invests in discretionary long/short equities” and recently committed $250 million to Quantopian, an algorithmic trading platform and hedge fund where quants develop and analyze models and receive a performance fee if they are used by the fund and generate profits.

Can computers and machine learning outperform human experts in long/short equity trading?

Machine learning can handle large volumes of data and many inputs, also known as features, attributes or predictors. These can include technical and fundamental indicators. Unsupervised learning may be used to determine the most promising group of features and then supervised classification on a training set followed by validation on a test set can be used to find a model. But what are the chances of success of this process?

Due to the low signal-to-noise ratio in financial prices series, the effectiveness of machine learning depends on the quality of the features used. Machine learning algos, no matter how advanced, cannot and will not find gold where there is none. Machine leaning can identify the best way of collecting the gold at the lowest possible risk provided that the gold is there in the first place.

Most machine learning long/short equity implementations use common features and classification algorithms and for this reason they cannot generate alpha. Features must be as unique as possible and should contribute towards identification of anomalies that are exploited fast and before other market participants. Otherwise profit potential diminishes and at some point it may turn negative.

However, the shift to machine leaning may be also motivated by dismal performance results in the last six years, as shown in the chart below.

According to BarclayHedge, long/short equity hedge funds have consistently underperformed the S&P 500 total return since 2012.

One reason for the underperformance may be crowded trades since most hedge funds use similar discretionary models. Another reason is that Fed policies may have adversely affected the ability of market neutral strategies to capture alpha. There are possibly multiple factors that have contributed to the low performance of long/short equity.

However, in my opinion, the most important reason for the inability of these models to provide higher returns is that there are not enough counterparties to take the other side and lose, i.e., uninformed and random human technical traders. Trading is nowadays 85% algos since most human traders are already driven out of the markets.

Therefore, the counterparties are now the algos and machine learning tries to fool them into taking the opposite side of the trade. This has been called a Battle of Machines. Traders and funds with the best machines win.

If you have any questions or comments, happy to connect on Twitter: @mikeharrisNY

About the author: Michael Harris is a trader and best selling author. He is also the developer of the first commercial software for identifying parameter-less patterns in price action 17 years ago. In the last seven years he has worked on the development of DLPAL, a software program that can be used to identify short-term anomalies in market data for use with fixed and machine learning models. Click here for more.

Disclaimer: The information in this article reflects the author’s opinions and is provided for informational purposes only. None of the information contained in this article constitutes a recommendation that any particular security, portfolio of securities, transaction, or investment strategy is suitable for any specific person. Read the full disclaimer here.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.