Less Guts, More Gain
Given that it’s well established that people are behaviourally biased and that market prices are impacted by these biases you might think that it ought to be possible to profitably trade on the underlying anomalies that these generate. However, by and large, it seems that this doesn’t happen, either because the anomaly vanishes as soon as it becomes widely known or because it’s not actually possible to trade on the thing for one technical reason or another.
If this is the case then behavioral finance ends up being an interesting research area but one that’s difficult, at least for ordinary mortals, to exploit other than through the tried and trusted means of learning to ignore the instincts of our guts. Still, as you might expect there are people out there seriously looking at whether there are profitable trading strategies that might be exploited. The slightly surprising answer seems to be that there are.
Vanishing Anomalies
The various anomalies that have been spotted down the years tend to fall into a number of categories. There are calendar effects, such as the January Effect, which seem to be related to waves of correlated human behaviour triggered by temporal events and there are predictive variable effects related to anomalous stock returns such as price/earnings ratios, price to book ratios, etc. A sub-set of these are mean reversion effects, such as those demonstrated by Thaler and De Bondt, where underperforming stocks tend to outperform and vice versa and also the opposite of mean reversion, momentum effects, where recent performance is predictive of future performance.
Trying to pick winning strategies out of this mess of pottage is a challenge in its own right. As we’ve previously seen William Schwert has demonstrated, at least partially, that as soon as any anomaly gets a significant following it tends to vanish, probably because of investors attempts to pre-empt it. In essence, some investors slightly loosen their buying criteria to get ahead of the pack, which results in buying or selling pressure in the opposite direction to that predicted and the anomaly dissolves in noise.
Because of this problem a number of researchers have wondered whether an old-fashioned diversification strategy across anomalies or some kind of backward looking anomaly timing algorithm might generate better and more reliable results. The results appear to suggest that they do, although what happens when investors try to implement these strategies is anyone’s guess.
Diversifying Anomalies
In Diversification Across Characteristics the researcher, Eric Hjalmarsson, has looked at what the effects would be of equal weight diversification across a range of anomalies and has found that the net result is significantly more profitable than a single characteristic portfolio. He looked at short and long term reversals, medium term momentum, book to market ratios, cashflow price ratio, earnings price ratio and corporation size.
Implementing a long-short portfolio with equal weights across all characteristics and back-testing to 1951 this strongly outperformed most of the individual characteristic portfolios over all time periods. Notably, over the last ten years of dismal market performance, this strategy still worked well even though many of the individual portfolios didn’t. Many, although not all, of the selected characteristics are weakly correlated – the exception being the three valuation ratios, which are negative correlated with short-term reversals and only weakly correlated with momentum.
Overall it looks like the combined benefits of exploiting these pricing anomalies, combined with the diversification benefits of using multiple, often uncorrelated characteristics, generates a significant overall benefit for investors. Whether this would hold going forward, though, is another matter as frequently strategies that work well in backtesting fail going forward.
Data Snooping
This failure isn’t simply a question of investors trying to exploit an anomaly and destroying it, it’s also often caused by a illusion of the data set which researchers fail to guard against known as data snooping bias. The problem is that in any large data set there will be any number of correlations which are statistically significant but also entirely random. A statistical significance test is actually a probability that an apparently significant finding arose purely by chance – and in a large data set you have to deal with the realistic chances that you might happen upon one of these.
It’s unlikely that the previous research would suffer from this, being in essence a study of studies – although to be sure you’d need to look at the underlying data – but data snooping bias is a constant danger for researchers in the financial arena and it’s important to look at how they deal with this possibility. One interesting approach, which also lends itself to the theme of investing based around the use of multiple anomalies, has been developed by Zhijian Huang, who’s investigated whether a profitable trading strategy can be developed by flipping your investing strategy between anomalies.
Rebalancing Anomalies
In Real-Time Profitability of Published Anomalies he’s tried to recreate the historical sequence of trading anomaly discoveries and then backtested to see whether choosing between the known universe of characteristics at any given time could have yielded abnormal returns. He did this by trawling back through the most prestigious financial research journals to identify when the various anomalies became widely known to the general investing populace.
This, of course, is a neat idea because he’s not back-testing on the anomalies before they were known so the study only uses data from published and therefore exploitable anomalies. Hence, if these are impacted by irrational investors all suddenly trying to use them this will show up in his results. The trading strategy proposed is a simple one – the investor selects a “training period” and then looks for the anomaly that generated the best returns in that period and then uses it for a defined holding period before rebalancing on whichever anomaly is now the most successful.
Abnormal Anomalies
It’s hard to believe that such a simple strategy could generate excess returns so, naturally, it does:
More Guts, More Gains
Overall it’s a nicely balanced paper – the findings are interesting and may be exploitable but they’re still subject to the vagaries of human psychology. As the researcher points out for one of his scenarios 1999, 2001 and 2002 ranked among the best years and 1998 and 2000 the worst:
Or are there?
Related Articles: Pricing Anomalies: Now You See Me, Now You Don't, Value in Mean Reversion, Rock On, January Effect
Given that it’s well established that people are behaviourally biased and that market prices are impacted by these biases you might think that it ought to be possible to profitably trade on the underlying anomalies that these generate. However, by and large, it seems that this doesn’t happen, either because the anomaly vanishes as soon as it becomes widely known or because it’s not actually possible to trade on the thing for one technical reason or another.
If this is the case then behavioral finance ends up being an interesting research area but one that’s difficult, at least for ordinary mortals, to exploit other than through the tried and trusted means of learning to ignore the instincts of our guts. Still, as you might expect there are people out there seriously looking at whether there are profitable trading strategies that might be exploited. The slightly surprising answer seems to be that there are.
Vanishing Anomalies
The various anomalies that have been spotted down the years tend to fall into a number of categories. There are calendar effects, such as the January Effect, which seem to be related to waves of correlated human behaviour triggered by temporal events and there are predictive variable effects related to anomalous stock returns such as price/earnings ratios, price to book ratios, etc. A sub-set of these are mean reversion effects, such as those demonstrated by Thaler and De Bondt, where underperforming stocks tend to outperform and vice versa and also the opposite of mean reversion, momentum effects, where recent performance is predictive of future performance.
Trying to pick winning strategies out of this mess of pottage is a challenge in its own right. As we’ve previously seen William Schwert has demonstrated, at least partially, that as soon as any anomaly gets a significant following it tends to vanish, probably because of investors attempts to pre-empt it. In essence, some investors slightly loosen their buying criteria to get ahead of the pack, which results in buying or selling pressure in the opposite direction to that predicted and the anomaly dissolves in noise.
Because of this problem a number of researchers have wondered whether an old-fashioned diversification strategy across anomalies or some kind of backward looking anomaly timing algorithm might generate better and more reliable results. The results appear to suggest that they do, although what happens when investors try to implement these strategies is anyone’s guess.
Diversifying Anomalies
In Diversification Across Characteristics the researcher, Eric Hjalmarsson, has looked at what the effects would be of equal weight diversification across a range of anomalies and has found that the net result is significantly more profitable than a single characteristic portfolio. He looked at short and long term reversals, medium term momentum, book to market ratios, cashflow price ratio, earnings price ratio and corporation size.
Implementing a long-short portfolio with equal weights across all characteristics and back-testing to 1951 this strongly outperformed most of the individual characteristic portfolios over all time periods. Notably, over the last ten years of dismal market performance, this strategy still worked well even though many of the individual portfolios didn’t. Many, although not all, of the selected characteristics are weakly correlated – the exception being the three valuation ratios, which are negative correlated with short-term reversals and only weakly correlated with momentum.
Overall it looks like the combined benefits of exploiting these pricing anomalies, combined with the diversification benefits of using multiple, often uncorrelated characteristics, generates a significant overall benefit for investors. Whether this would hold going forward, though, is another matter as frequently strategies that work well in backtesting fail going forward.
Data Snooping
This failure isn’t simply a question of investors trying to exploit an anomaly and destroying it, it’s also often caused by a illusion of the data set which researchers fail to guard against known as data snooping bias. The problem is that in any large data set there will be any number of correlations which are statistically significant but also entirely random. A statistical significance test is actually a probability that an apparently significant finding arose purely by chance – and in a large data set you have to deal with the realistic chances that you might happen upon one of these.
It’s unlikely that the previous research would suffer from this, being in essence a study of studies – although to be sure you’d need to look at the underlying data – but data snooping bias is a constant danger for researchers in the financial arena and it’s important to look at how they deal with this possibility. One interesting approach, which also lends itself to the theme of investing based around the use of multiple anomalies, has been developed by Zhijian Huang, who’s investigated whether a profitable trading strategy can be developed by flipping your investing strategy between anomalies.
Rebalancing Anomalies
In Real-Time Profitability of Published Anomalies he’s tried to recreate the historical sequence of trading anomaly discoveries and then backtested to see whether choosing between the known universe of characteristics at any given time could have yielded abnormal returns. He did this by trawling back through the most prestigious financial research journals to identify when the various anomalies became widely known to the general investing populace.
This, of course, is a neat idea because he’s not back-testing on the anomalies before they were known so the study only uses data from published and therefore exploitable anomalies. Hence, if these are impacted by irrational investors all suddenly trying to use them this will show up in his results. The trading strategy proposed is a simple one – the investor selects a “training period” and then looks for the anomaly that generated the best returns in that period and then uses it for a defined holding period before rebalancing on whichever anomaly is now the most successful.
Abnormal Anomalies
It’s hard to believe that such a simple strategy could generate excess returns so, naturally, it does:
“The average annual excess return over the buy-and-hold return of a market benchmark ranges from 4.76% to 12.25% and is significant at the 5% confidence level under all scenarios. If the real-time trader happened to choose two years or five years as the training length, the performance would be even better with excess returns and Alpha's significant at the 1% level for these six cases.”Which is, frankly, remarkable. The study also attempts to include trading costs and shows that for all training periods greater than a year these don’t make any appreciable difference. Huang also looked at which anomalies ended up getting chosen most frequently. It turns out that the January effect and momentum effects are those with the strongest impact on the results although if these are removed from the universe of available characteristics the results are still positive.
More Guts, More Gains
Overall it’s a nicely balanced paper – the findings are interesting and may be exploitable but they’re still subject to the vagaries of human psychology. As the researcher points out for one of his scenarios 1999, 2001 and 2002 ranked among the best years and 1998 and 2000 the worst:
“If the real-time trader or his mutual fund clients learn from recent performance, the trader would miss the best year of 1999 because of the bad performance in 1998. If he corrects this mistake after observing the good performance in 1999, he would run into another tough year of 2000. Based on the past year performance, the real-time trader would not pick the right side until 2002 when both the previous and the current performances are good. However, would the real-time trader (or his investors) change this learning behavior at the end of 2001 after being fooled by past performances for three consecutive years?”Sometimes it’s not enough to have a good strategy, you also need to understand why it’s a good strategy in order to profit in the long term. Automated anomaly exploitation can be extremely difficult, psychologically, for this very reason. As usual, with markets, there are no free lunches.
Or are there?
Related Articles: Pricing Anomalies: Now You See Me, Now You Don't, Value in Mean Reversion, Rock On, January Effect
Sometimes it’s not enough to have a good strategy, you also need to understand why it’s a good strategy in order to profit in the long term.
ReplyDeleteI think this is key. You have to have strong confidence in your strategy or you will not stick with it. And, if you don't stick with it, it will not work.
There are free lunches. But there aren't many.
Fortunately, it's like finding a marriage partner. You only need to make one good call and you win the game for life.
Rob