Contrarian Investing

I was inspired by this article to make some comments about investing 'techniques.' To summarize the article, the guy argues that investment techniques like technical analysis (using past price data to forecast future price movements), fundamental analysis (a la Graham, Buffett, etc.; also called value investing), and something called Elliott Wave (no idea) don't work. What he means by 'don't work' is that these methods don't offer reliable systems for generating abnormal returns regularly and reliably. That's fine; that's what Efficient Market Hypothesis says: you can't regularly beat the market, although you can sometimes.

Let me quickly say one thing about value investing, since that's an area of research of mine. It does appear to be the case the value investing methods (i.e. identifying underpriced stocks) generates abnormal returns over time. Why? Stocks become underpriced because of analyst disregard and investor ignorance. Those stocks offer good opportunities for capital gains, but they also tie up a lot of liquidity because price recovery tends to be sluggish. So mutual funds and institutions that need the ability to sell quickly avoid such stocks, making the market for value stocks rather inefficient.

Anyway, the author goes on to discuss the investment technique that he thinks is 100% reliable. He doesn't appear to be selling anything, so I don't doubt that he believes his claim to be true. And it is quite a plausible technique: be a contrarian.

What does it mean to be a contrarian? It means going against the crowd. In investing, it means selling when everyone is buying, and buying when everyone is selling. Except that in the stock market, for every buyer there's a seller, so what it really means is buy when prices are low, and sell when prices are high. Hey, wait, isn't that just the old saw of 'buy low, sell high?' Well, yes it is. And that's really the genius of such a rule. It's old wisdom that nobody follows. Because investors display herding behavior, most of them end up buying high and selling low. The bond buying frenzy of a couple years ago immediately springs to mind. If you're chasing returns, you're pouring money down the drain.

A somewhat more refined view is to look at investor sentiment. Figure out what the herd thinks will happen, and then do the opposite. In this refinement, one would take the view opposing what most people think will happen. If the herd thinks stock prices will be up next week, take a short position, and vice-versa. This technique was articulated in the finance literature by DeBondt and Thaler (1985) and termed 'reversal.' Reversal can exist in the short-run (weekly) or long-run (many months). Reversal has always been suspect, since it is behavioral then the one thing you can trust is for people to be unpredictable just when you want them to be predictable.

So, with this view in mind, I ran some econometrics. Cause that's how I roll. I used the investor sentiment measure (a weekly poll) produced by the American Association of Individual Investors (AAII). They ask people if they are bullish, bearish, or neutral about the market, and then the data is the percentage of total respondents that claim of these three. I use the spread between bullish and bearish as my sentiment indicator. My dependent variable is the weekly return on the S&P 500. I check my work with monthly data afterwards. I run the following model:

Return next week = A*Return this week + B*Spread this week + Error

Spread next week = C*Return this week + D*Spread this week + Error

It's a little technical, but this is a simultaneous equations model of time series (called a vector autoregression or VAR). Essentially it can tell us if the spread has any correlation with the return in the week following, or if the return has any correlation with next week's spread. I go back as far as four weeks (four months to check my work). I find the spread this week is not at all correlated with next week's return. Same with monthly data. Past returns are strongly correlated with the spread, though. If the stock market's been doing well, then people feel it will continue to do well, and vice versa. And the spread itself is highly persistent - the herd doesn't change its mind very quickly or often. 

I refine the measure a bit to check if extremes matter more. So I impose, along with the spread measurement, a measure of when the spread is in the top bullish quartile or top bearish quartile. In other words, is there any predictive ability of the spread when it is quite high or low? Some, but not really. Next week's returns will be higher if the spread four weeks ago was abnormally bearish. So if everyone is bearish, and I buy, then I'll do better in four weeks. But not in the interim. Sounds iffy to me. I find this same result for months, but if the week thing is correct it should appear in the lagged month, not the lagged four months. So I think this is a feature of the data (sunspots), not something we can use to predict the market.

Now, this isn't evidence that disproves this fellow's investing technique. It just doesn't offer any evidence in favor of it. I'd be happy to try different sentiment measures. Incidentally, I did try the University of Michigan Consumer Confidence Index. That's no better than AAII's measure. So I'm not willing, at this point, to give up on old-fashioned Graham.