Turning Points

History verifies that the stock market rises higher than expected and falls lower than expected. Our investigations of Tobin’s q and stock market returns tell us more. They tell us about the turning points.

Background

From Tobin q Errors:

“Error magnitudes (i.e., absolute values) were almost independent of actual returns. Interestingly, the smallest error magnitudes were associated with the percentage earnings yield 100E10/P.”

“The errors themselves had a negative correlation with actual returns. When the actual returns were small, the errors were positive and larger than normal. In such cases, the predicted returns were bigger than the actual returns. When the actual returns turned out badly, the actual returns were worse (smaller) than predicted.”

“When the actual returns were high, the errors were negative. When the actual returns were most favorable, the actual returns turned out better (larger) than predicted.”

“For a purely random error, we would not expect any such relationship. Because error magnitudes were independent, I conclude that the middle section with the bulk of the data defines the regression equations. All of the projections have problems with the extremes: the very best and worst returns.”

Implications

The prediction errors occur because of turning points, when valuations reverse course. Direct predictions based on valuations (both for 100E10/P and all six variants of Tobin’s q that I have examined) make no directional adjustment. They do not take into account whether multiples have been increasing or decreasing.

Starting from 2000 with P/E10=44: the predicted year 2010 return is -1.1% per year plus and minus 6% at the outer confidence limits (5% and 95%, and roughly plus and minus 3% at the 20% and 80% confidence levels).

Starting from 2000 with P/E10=44: the predicted year 2020 return is 1.0% per year plus and minus 4% at the outer confidence limits (5% and 95%, and roughly plus and minus 2% at the 20% and 80% confidence levels).

Based on what we have learned about error magnitudes when predicted returns are low, we expect that the actual results will be worse. From Tobin q Survey Follow-On, the historical “low return outlier is minus 7% on the downside” when using x=100E10/P. The typical low return adjustments using 100E10/P predictions are minus 6% at Year 10 and minus 3% at Years 15 and 20.

Starting from today with P/E10=26: the predicted year 2016 expected return is 1.3% plus and minus 6% at the outer confidence limits (5% and 95%, and roughly plus and minus 3% at the 20% and 80% confidence levels).

These returns are low, but they are not at a turning point. There is no need for an adjustment.

Balanced against this is the memory effect. It was mentioned in Rational Pessimism: Predicting Equity Returns using Tobin's q and Price/Earnings Ratios. The 1990 ten-year real return was 14.55%, 7% to 8% above the long-term annualized, total return of 6.8%. This tells us to increase our 2010 prediction. The memory effect has little or no influence on either the 2016 prediction, since any adjustment will be close to zero, or the 2020 prediction, because the timeframe is too long.

So what will the 2010 return be as measured from 2000?

As of 2005, the stock market had lost -4.6% per year (real, annualized, total return) starting from January 2000. If the stock market were to continue sideways between 2005 and 2010, the annualized loss percentage would be -2.3% per year (real, annualized, total return).

Although not conclusive, this suggests that turning points dominate the memory effect.

Summary

Measured from today, we expect the 10-year prediction to be accurate. The expected return for 2016 is 1.3% plus and minus 6% at the outer confidence limits (5% and 95%, and roughly plus and minus 3% at the 20% and 80% confidence levels).

Measured from 2000, we expect the 20-year prediction to be accurate. The year 2020 return will be 1.0% plus and minus 4% at the outer confidence limits (5% and 95%, and roughly plus and minus 2% at the 20% and 80% confidence levels).

The estimate for 2010 is much less certain.

Relative to January 2000, the Stock-Return Predictor estimates the year 2010 return will be -1.1% plus and minus 6% at the outer confidence limits (5% and 95%, and roughly plus and minus 3% at the 20% and 80% confidence levels). Based on what we have learned about error magnitudes when predicting low returns, we anticipate that the actual results will be worse.

Balanced against this is the memory effect. The memory effect implies higher returns. But judging from the returns from 2000 through 2005, the turning point effect will have the greater influence.

Have fun.

John Walter Russell
August 14, 2006