I have a confession to make. I’m addicted to market indicators. Sentiment, valuation, insider behavior, even the odd moving average. Mind you, I take them all with a barrel of salt, but I can’t resist the things.

My all-time favorite is equity mutual-fund cash. Over the past decade, I’ve been impressed with the coincidence of high fund cash with market bottoms, so when I came across a tabulation of this indicator back to 1970 on the Investment Company Institute site, I just couldn’t resist. The data are striking—record low cash levels just before the 1973-74 debacle, and record high levels right at the bottoms in 1984, 1987, and late 1990. I greedily downloaded, pasted, and correlated, and hey, presto, came up with the following plot of fund cash versus forward 3-year annualized returns:

There’s a pretty impressive upward slope to the data cloud, and in fact the correlation coefficient of the quarterly data points is 0.56. In other words, you can explain 31 percent (0.56 squared) of future returns with this indicator. For those of you unfamiliar with returns data, that’s about as good as it gets short of a microphone in Alan Greenspan’s living room. I’m not aware of any indicator that comes even close during the past three decades. The traditional valuation parameters—price-to-book, P/E, and dividend yield—which have done so well since 1900, have fallen flat on their faces since 1994. And contrary to what you’ve heard, bullish or bearish sentiment among investors and newsletter writers is useless for predicting future returns.

Surely, had you had the foresight to tailor your equity exposure according to fund cash level, you’d have beaten the market or at least lessened your risk level by avoiding the rough patches…

So, I began formulating simple trading rules. The first is, an all-or-none policy of 100 percent S&P 500 above a given cash level or 100 percent Treasury bills below it, adjusted on a quarterly basis. In other words, if the filter was set at 6 percent, then 100 percent S&P 500 was held for the quarter when fund cash was above that level, and 100 percent Treasury bills were held when fund cash was below that level. Here are the results for the 28-year period from January 1970 to September 1998:

Fund Cash "Filter" (percent)	Annualized Return 1970-98	Standard Deviation 1970-98	Sharpe Ratio	% Periods Holding Stock
0 to 4	12.77	16.19	.370	100
5	12.93	15.94	.385	89.6
6	11.73	15.53	.319	83.5
7	11.33	14.86	.306	73.9
8	11.44	13.57	.343	59.1
8.5	12.03	13.30	.395	48.0
9	10.22	11.81	.292	37.8
10	7.89	8.24	.135	17.4
11	7.01	2.24	.103	2.6
12	7.19	2.05	.200	1.7

For starters, the first row simply shows that since the lowest cash position was 4.1 percent (April 1972), setting the filter below that value resulted in a 100-percent stock portfolio for the entire period, with a return of 12.77 percent, a standard deviation of 16.19 percent, and a Sharpe ratio of .370. The key number here is the Sharpe ratio (calculated as [return-T bill return]/SD, where the T-bill return for the period was 6.78 percent) This measures risk-adjusted return.

Notice that the only filters which result in superior risk-adjusted return are discrete values of 5 percent (Sharpe ratio of 0.385) and 8.5 percent (Sharpe ratio of 0.395). And, neither Sharpe ratio is much higher than the 100 percent S&P 500 buy-and-hold one. In fact, lowering or raising the filter slightly, say to 8.4 or 8.6 percent, results in Sharpe ratios significantly less than the buy-and-hold value. Such are the vagaries of a system which requires switching back and forth between 100 percent cash and 100 percent equity—something which only newsletter writers and a few of their more gullible readers seem willing to do.

OK, you say, suppose we consider a more reasonable system—one in which we start with a "policy" equity exposure which we modify in "scaled" fashion according to cash level. Let’s start with a "policy" mix of 60 percent stock and 40 percent T-bills. This results in a return of 10.73 percent for the 1970-98 period with a standard deviation of 9.68 percent and a Sharpe ratio of .408.

Now let’s suppose an algorithm was established which allowed one to raise or lower the equity exposure according to a cash level "thermostat," above or below which equity was lowered or raised, and a "multiplier," which was used to calculate how sensitive the change in exposure was to be to such cash level changes.

For example, assume the "thermostat" was set at 8 percent fund cash and the "multiplier" was set at a value of 10. In a quarter when fund cash was actually 5 percent, then equity exposure was lowered by (8-5)*10, or 30 percent. If the fund cash level for the quarter was 9.5 percent, then equity exposure was raised by (9.5-8)*10, or 15 percent.

I found that in all cases in which the multiplier was positive, Sharpe ratios were lowered using this algorithm. In fact, the only way that the efficiency of the portfolios could be raised (and then only slightly) was with the use of small negative multipliers and a "thermostat" in the region of 1 percent—something which no rational portfolio strategist would do. (In this bizarre case one always holds less equity than "policy," since fund cash is always more than the "thermostat," and the multiplier is negative. Further, the negative multiplier means that one increases stock exposure when fund cash decreases! If you find this counterintuitive, you’re not alone.)

If you'd like to play with this algorithm yourself, click here for the self-executing zipped Excel file and text instructions.

The take-home lesson of this rather frustrating exercise is this: The next time you hear some guru wearing an Armani suit telling Uncle Lou about what his sure-fire indicator shows this week, remember these two things:

1. He’s likely not had the courage (or worse, doesn’t know how) to calculate a correlation with past returns. If he did, the correlation would probably be much less than the .56 value for equity-fund cash.
   
   
2. Even if it were as high as the fund-cash indicator, it still won’t make you a risk-adjusted dime.

The reason why even the best predictors of future returns known to mankind do not improve portfolio efficiency is simple: Since 1926 stock prices have risen in two out of three years. For any timing system to succeed, it must therefore supply correct calls 70 percent of the time. Even the proverbial microphone in the chairman’s townhouse isn’t that good.

My head will still get turned by every pretty market indicator I see, but if I make any portfolio changes because of them, they will be small and infrequent.

copyright (c) 1999, William J. Bernstein