Close

    Newsletter Sentiment: It's the Degree of Agreement That Counts

    by Mark Hulbert

    Many contrarians for a number of decades have been basing their market-timing judgments on the consensus opinion of investment newsletter editors. They claim great success by turning bearish whenever that consensus reaches bullish extremes—and turning bullish as it reaches a bearish extreme.

    Nevertheless, academic researchers have been unable to discover any statistical significance to market timing systems based on such data. The most comprehensive study, by Duke University’s John Graham and Campbell Harvey (“Market Timing Ability and Volatility Implied in Investment Newsletters’ Asset Allocation Recommendations,” Journal of Financial Economics, 1996), was unable to locate any mechanical decision rule based on the consensus newsletter recommendation that beat a buy and hold.

    Recently, however, researchers have discovered another way of using newsletter sentiment data that does appear to have statistical significance. Rather than concentrate on the average recommendation among all newsletter editors, this new research has focused on the degree of dispersion among their various recommendations. This dispersion appears to be correlated with any of several market phenomena.

    The researchers who focused on the dispersion of newsletter editors’ forecasts were William Goetzmann of the Yale School of Management and Massimo Massa of France’s INSEAD University (“Index Funds and Stock Market Growth,” a working paper). These two researchers were studying the various factors causing mutual fund inflows and outflows, and were curious if the forecasts of investment newsletters had any impact. While they found that the consensus newsletter opinion had some explanatory power, they found that they could explain even more of the inflows and outflows by focusing on the dispersion of opinion among newsletters. They concluded that “a higher dispersion of newsletter opinions lowers the inflows and increases the outflows.”

    The HFD Study

    This finding prompted the Hulbert Financial Digest (HFD) to measure other ways in which dispersion of newsletters’ market-timing opinions may contain useful information. We were particularly interested in discovering whether this dispersion was correlated with movements in the stock market as a whole. Given the results of the Goetzmann/Massa study, we suspected that it would be, since mutual fund inflows may cause the stock market to go up, and vice versa.

    Figure 1.
    Change in Wilshire 5000
    vs. Dispersion of Newsletter
    Equity Exposure
    Recommendations
    CLICK ON IMAGE TO
    SEE FULL SIZE.

    To find out, we calculated the dispersion of newsletters’ percentage exposure to the stock market. The statistic we used was standard deviation, which measures the amount by which newsletters’ recommended market exposure percentages varied around the average recommendation. That means that the higher the standard deviation, the greater the dispersion of opinion in terms of recommended stock market exposure.

    We examined these standard deviations in recommendations for every trading day since June 30, 1980, which is when the HFD began monitoring newsletter performance. That meant we had over 5,500 observations. For each of these days, furthermore, we measured the percentage change over the previous five and 20 trading sessions, as well as the percentage change in the Wilshire 5000 over the subsequent five and 20 trading sessions.

    While there was a wide range of standard deviations over these 22 years, most fell between 40% and 60%. Only 133 of the trading days over the last 22 years had standard deviations below 40%, and only 119 above 60%. The average standard deviation was 48.6%.

    The correlations between these standard deviations and subsequent market moves were significant, as seen in Figure 1. Lower standard deviations—which meant more agreement in market exposure among the newsletters—were associated with stronger market rallies than were higher standard deviations. For example, over the five and 20 trading sessions following days on which the standard deviation was below 40%, the market gained 70 and 192 basis points (a basis point is 0.01%), respectively. Following days on which the standard deviation was above 60%—meaning less agreement in market exposure among the newsletters—the market dropped 57 basis points over the five-day trading sessions and 153 basis points over the 20-day trading sessions. That’s a difference over a five-trading-day period of 127 basis points and over a 20-trading-day period of 345 basis points. (By the way, these results were not skewed by the 1987 crash; none of the days in which the standard deviation was above 60% came during October 1987.)

    In addition to correlating the raw standard deviations with the market’s subsequent movements, we also correlated percentage changes in the standard deviation to the market’s subsequent performance. Here too there were strong correlations.

    Why This Pattern?

    A plausible hypothesis is available to explain at least part of these correlations’ existence: Uncertainty on the part of professional investment advisers causes investors to take some money off the table. Insofar as this hypothesis is correct, uncertainty should be a more reliable predictor of below-market performance than relative certainty is a predictor of market rallies. This does appear to be the case from the results of the study: The findings are more significant in the instances where there is more dispersion in the newsletter recommendations (the highest standard deviations).

    Further confirmation comes from other research conducted by Goetzmann and Massa. They found that the levels of disagreement among futures traders, as measured by the open interest, also were significantly correlated with mutual fund flows. As with the HFD’s results, high levels of disagreement were better predictors of fund outflows than low levels of disagreement were of inflows.

    Nevertheless, questions still remain for further research. The main one is why low dispersion among advisers’ opinions should in and of itself be a predictor of market rallies. After all, that low dispersion might represent a consensus that the market is about to decline. Why should that be as good a predictor of a market rally as a consensus that the market is about to go up?

    Where We Stand Now

    One thing that the HFD has not yet studied is whether dispersion of advisory opinion can be the basis of a market-timing strategy that beats a buy-and-hold strategy. Transaction costs and taxes doom many a strategy that looks promising on paper. That fate also may await a strategy based on dispersion of opinion. Those costs would be particularly high if trades were made over shorter time periods.

    But even if transaction costs and taxes end up dooming such a strategy, or even if you are not a market timer and have no interest in the market’s short-term gyrations, the findings of this new HFD study are relevant. They help to reveal the market’s internal workings, and provide yet more evidence of the degree to which behavioral factors influence the stock market. Investors ignore them at their peril.

    The market, it appears, likes consensus.


    Mark Hulbert is editor of the Hulbert Financial Digest, a newsletter that ranks the performance of investment advisory newsletters. It is published monthly and is located at 5051B Backlick Rd., Annandale, Va. 22003; 703/750-9060; www.hulbertdigest.com.

    This column appears quarterly and is copyrighted by HFD and AAII.


→ Mark Hulbert