In his 2008 book, Short Term Trading Strategies That Work, Larry Connors, CEO and Founder of The Connors Group, shares “more than two decades of research and trading knowledge.” He states in the introductory chapter:
“Philosophically, I live in the world of reversion to the mean when it comes to trading. What that simply means is that something stretched too far will snap back. I didn’t come up with that idea. It’s been around for decades. What I have done though is an attempt to quantify it.“
Most of the quantifying tests use daily data spanning 1995-2007 for a large sample of stocks and the S&P 500 index. Based on past reviews of hundreds of anomaly studies, here are a few observations on the analyses presented in this book:
In general, Short Term Trading Strategies That Work is for frequent traders who expect to take some losses but come out ahead on average over many trades. The book presents statistical results of round-trip trading anomalies simply and clearly. Readers should be able to replicate and explore some conclusions, such as those focused on the S&P 500 index and the Chicago Board Options Exchange Volatility Index (VIX). Few readers are able to replicate conclusions based on the “8 million trades of stocks from 1995-2007” cited in several analyses in the book.
Some of his empirically derived rules include: buy pullbacks (not breakouts), buy above (not below) 200-day moving averages, buy fear (not greed), do not use stops, hold overnight, hold near the end of the month, and use a 2-period Relative Strength Index.
Some of the analyses in the book establish robustness via systematic variation (varying the input systematically affects the output). Some of the analyses examine dependence of results on market state (by adding price relative to long-term moving average as a separate condition). However, other tests of robustness and economic value of discovered anomalies are apparently lacking, as follows:
The analyses apparently do not take data snooping bias into account. Data mining bias accrues as researchers search for anomalies by testing more and more rules or rule variations on the same (or a highly correlated) set of data, increasing the probability that they will discover “luck” that does not persist out of sample. There can be a “second hand smoke” effect that obscures the extent of data mining when researchers borrow or build upon results of others without knowing how many rules the others considered. To the extent that data mining bias is present but uncorrected, statistical results tend to overstate out-of-sample experience. See the synopses of Chapter 6 and Chapter 8 in our review of Evidence-Based Technical Analysis by David Aronson for additional thoughts on data mining bias. The analyses in Short Term Trading Strategies That Work apparently do not correct for data mining bias in statistical results or employ out-of-sample testing to obviate the effect of data mining bias.
The analyses focus on average returns for anomalies, but most do not explore variability of returns and consequent potential drawdowns. This concern relates to assumptions about the distributions of the samples underlying the averages and the risk of “Black Swan” wipeouts.
Most of the analyses do not present subperiod testing that might verify the reliability and persistence of anomalies. Even when an anomaly appears solid for an entire sample period, subperiod testing sometimes shows that there are fairly long intervals when the anomaly does not work or that the anomaly is declining in strength over time (perhaps due to publication and widening use).
The analyses generally do not explore the degree to which trades from large samples (such as the cited “8 million trades of stocks from 1995-2007”) are independently exploitable or continuously available. If a rule generates many overlapping trading opportunities (e.g., hundreds of stocks giving Relative Strength Index signals during the same week), a trader with limited capital could not trade them all. These concentrated, overlapping opportunities may drive the attractiveness of overall sample statistics. Conversely, a rule might not generate any useful signals for months at a time. Overall statistics from such unevenly distributed opportunities may be strong but still not translate into a practicable trading strategy.
The analyses are generally hypothetical and do not explore the sensitivity of profitability to trading frictions.
In summary, equity traders may find the trading rules in Short Term Trading Strategies That Work interesting, but they should consider potential limitations in the supporting analyses and recognize the challenge of reliably extracting economic value from such rules with trading frictions in a real, continuously managed series of trades.
Frequently Asked Questions
What is mean reversion in short-term stock trading?
Mean reversion in short-term stock trading is the idea that something stretched too far, such as a price that has moved sharply away from its norm, will snap back. Larry Connors, in his 2008 book Short Term Trading Strategies That Work, says the idea has existed for decades and that his contribution is an attempt to quantify it using daily data from 1995-2007 on a large sample of stocks and the S&P 500 index.
Do short-term mean reversion rules like buying pullbacks and using a 2-period RSI actually work?
Short-term mean reversion rules such as buying pullbacks and using a 2-period RSI show positive average returns in Larry Connors' backtests on daily data from 1995-2007. However, the analyses are hypothetical, apparently do not correct for data snooping bias, and do not test out of sample, so the results likely overstate what traders could expect in real trading. The rules are suited to frequent traders who accept some losses but aim to come out ahead on average over many trades.
How are short-term trading rules like buying pullbacks or fear-based VIX signals tested?
Short-term trading rules like pullback buying and VIX-based signals are tested by measuring average returns of round-trip trades using daily data from 1995-2007 on a large sample of stocks and the S&P 500 index. Some analyses check robustness by systematically varying inputs, and some add price relative to the 200-day moving average as a market-state condition. Results based on the cited 8 million stock trades are hard for readers to replicate, while those on the S&P 500 and VIX are more replicable.
Is buying pullbacks the same as buying breakouts, and which do mean reversion rules favor?
Buying pullbacks is not the same as buying breakouts, and Larry Connors' mean reversion rules favor pullbacks over breakouts. His empirically derived rules also include buying above (not below) the 200-day moving average, buying fear (not greed), avoiding stops, holding overnight, and holding near the end of the month. These are findings from his hypothetical backtests and carry the data mining and trading friction caveats.
What are the limitations and risks of backtested short-term trading strategies like those from Larry Connors?
Backtested short-term strategies like Larry Connors' have several limitations: the analyses apparently do not correct for data mining bias or use out-of-sample testing, and most do not examine return variability, drawdowns, or Black Swan risk. Most also lack subperiod testing, do not check whether overlapping signals from a large sample (such as 8 million trades) are independently tradable, and do not test sensitivity to trading frictions. As a result, reliably extracting economic value from the rules in real, continuously managed trading is challenging.