Daily Speculations

January 2004

Sentimental Readings: Vic delves into recent academic offerings on the causes of crashes and investor sentiment.

Causes of Crashes: "Adaptive Expectations and Stock Market Crashes" has a nice model of a crash: "The rational traders observe a common signal that acts as a coordinating device. For certain values of this signal, they dump their shares. The resulting crash raises the naive trader's assessment of the risk in the market. Since naive traders are risk averse, they become less willing to own stocks. This lowers the market's risk-bearing capacity. Thus, the price remains lower for some time after the crash. Expecting this, rational traders have no incentive to bid up the stock price on the crash day. This makes the crash a self-fulfilling prophecy. " While the author (unknown) references Shiller and Summers favorably, and doubtless is part of the bearish meme of Prechter-Taleb-Buffett-Fleckenstein, it seems like a pretty good and rich description.

The Grossman and Zhou models, with symmetric information and two types of risk-averse investors who each maximize expected consumption utility. One type, the portfolio insurers. have an additional constraint that their wealth must not fall below a certain level. As fundamentals worsen, the portfolio insurers sell stock at an accelerating rate, leading to an increase in volatility." That sounds  pretty good also. --  Vic (Date: 01/15/2004 12:57:48

That Sentimental Feeling: Review of "Explaining Stock Market Correlation: A Gravity Model Approach," by Thomas Flavin, Margaret Hurley and Fabrice Rousseau (Natural University of Ireland). (9:35 a.m. Jan. 15) http://eprints.may.ie/archive/00000087/

In an effort sparked by the Baker and Wurgler paper (reviewed below) to come up with better measures of sentiment, I have been looking through the 40,000 or so studies of market sentiment. The gravity approach is one that apparently has been used to explain trade flows and relates to distance and size as key independent explanatory variables. The current authors explain daily market correlations over the 2000-2002 period for the 29 biggest markets. They conclude that the key determinants of the correlation are the greatest circular distance between the two, the product of the sizes, and their sharing of a common currency. They then expand the model to encapsulate some other key variables which turn out to be significant, including the number of overlapping hours the markets are open, the degree of industrialization in the market, the percentage of market capitalization accounted for by the five largest companies, the sharing of a common border, and the affinity of the legal system. This last is based on rule of law, danger of expropriation, corruption, efficiency, and strength of  contractual relations. Some countries with important industrial concentrations are Taiwan(47%), Korea ( 36%), India (30%), and Germany (26%). The U.S. is one of the least industrial, at 11%; the only country with less is Brazil at 6%. Table 4 in the paper gives a nice index of corporate governance for each of 29 companies classified by the four factors, with Mexico being the worst and Switzerland and Holland being the highest.
OK, not immediately useful, and their degree of overlap variable correlation is a part whole correlation, and no measures of the contributions to the total variation of each variable (just significance levels reported). But a workmanlike study that might lead one to develop some interesting insights. -- Vic

The Sentimental Investor: Malcolm Baker and Jeffrey Wurgler are familiar to our longtime readers as the young professors whose work formed one of the pillars of our optimistic forecast for 2003. Standing on their shoulders, we updated their seasoned stock-to-total capital ratio to form one of the pillars of our forecast of a 20% rise in the S&P last year (the other pillar being the much-maligned Fed Model, which we showed to be valid on a predictive basis despite the armchair quarterbacking of its defects). In a new paper, "Investor Sentiment and the Cross-Section of Stock Returns" (Dec. 3, 2003), they professors attempt to predict the kinds of stocks that will do well in a year, based on the investor sentiment in the previous year. They define sentiment with seven measures based on closed-end discount, NYSE volume turnover, number of IPOS, average first-day return of IPOs, percentage of equity in total capital raised, and the premium above book value received by companies paying dividends. They use yearly data on their independent variables, and over a 40-year period after many refinements and splits and novel definitions, they conclude that when sentiment is low, subsequent returns are higher on small stocks, high-volatility stocks, stocks of unprofitable companies, non-dividend-paying stocks, extreme growth stocks, and distressed stocks.

They explain this as an anomaly resulting from demand for the very uncertain and nebulous drying up during periods of low sentiment, and thus their prices get depressed. The professors created a composite index that strains predictive credibility, but the IPO premium and dividend premium correlate about 80% with it, so one can work with the best two independent variables with almost equal retrospective correlations, and better predictive work. (The authors believe that the closed-end fund discount is their best univariate indicator of sentiment.) The key correlations that the sentiments have on individual stocks is a size effect of about 1.5% a month, an age affect of about 1.44% a month, and a total risk effect of 1% a month. Presumably the results would be similar if one just took the market's change in the previous year and then bought risky stocks when markets were bad. Fortuitously the professors' data ends in 2001, so the terrible results I believe they would have achieved in 2002 and the greats results in 2003 are not in there. This is a very thought-provoking study, and aside from its use of hundreds of splits on 40 years of data, the work has legs. I don't know how the professors will react to my preliminary review, but will report their demurrals directly.

Baker and Wurgler pair are carving a niche with sharply focused articles pointing out defects and omissions in popular academic models, and then devising predictive yearly forecasts for the market and individual stocks based on the gaps between theory and practical wisdom. Their efforts are laudatory, except for their tendency to predict a very small number of future data points with an implicitly infinite set of independent variables and precious transformations thereupon.

Jeffrey Wurgler responds (Jan. 14-15):

The timing may be confusing. I think the following is correct. The Dec. 31, 1993, value of some sentiment proxy is used to predict monthly returns for each of the 12 months in 1994. But, for purposes of identifying a given stock's characteristics, we make sure to use only information that would have been available to investors at that point. Thus, a June 1994 stock return observation is classified using characteristics dated June 1993 and predicted with sentiment from the end of 1993, while a July 1994 stock return observation is classified with June 1994 characteristics, and also predicted with end of 1993 sentiment. The reason for all this mess is that the characteristics data is not available until FY end, which doesn't match calendar year end, and we wanted to ensure that accounting data not available to investors until, say, February, wasn't being used to predict the January return. (We took all this from Fama and French (1992).)

The annual data is available at the bottom of my webpage: www.stern.nyu.edu/~jwurgler

Monthly data might get you a tiny bit more, but I would guess very little, because the nature of the results is low-frequency booms and crashes. The way Malcolm and I think about the results is that when sentiment is high, there is say a 25% chance of a big crash in the next year. This crash will affect certain stocks much more than others, and this is what drives the
results. The other 75% of the time, though, sentiment may stay high or go even higher over the next year (I guess that is the nature of sentiment), and there may be little predictability.

--It is always easy to "predict" in the sample, especially when one is working with some nontrivial knowledge of market history. The real proof will be whether there is "out-of-sample" predictability. One could of course break up the 40-year period of our data into subperiods and look at each separately, but realistically our results are based on low-frequency bubbles and crashes, and there are only a few such market cycles within even a 40-year period. So, basically, only time will tell. In the meantime, one can gain some confidence from the fact that the results line up closely with intuition. When investor sentiment crashes, people avoid stocks with the perceived characteristics of "risk" in favor of those with the salient characteristics of "safety." This strikes us as a robust intuition and so it would be quite surprising if it did not continue to hold in the future."

"Who Cares About the Stock Market," a paper by Heloissa Marone, looks at the
Michigan Survey of consumers to see how the probability of feeling good is affected by  the stock market. Data is for 1990-2000 in the main. She finds that if the market increases by a few percentage in a month, the chances that survey respondents will feel better increases by about 2%. The effect is bigger for those who hold stocks. My goodness, what a part-whole fallacy, since financial well being is  coterminous with stock market wealth and respondents are answering truthfully. Nasdaq moves seemed to affect people's sense of well-being as much as the S&P during 1990s, and and the Dow not as much. Middle-age respondents were 3.3 times as sensitive to stock market moves as the old. "These results are consistent with the hypothesis that individuals dissave, or at least decrease the amount of wealth held in stocks , as they age". -- Vic