Most investors know that the US stock market is overvalued, as is often the case during the late stages of a bull market. This however doesn’t keep them from staying long and bullish, since short-term expectations are predominant. In the end, very few investors actually sell expensive markets and buy back in after they crash. This is not because buying low and selling high is technically difficult, but because it runs contrary to every instinct.
This is a perfect case for delegating decisions to an algorithm. As has been shown in many fields, a simple heuristic, strictly followed, will outperform the vast majority of experts. Here I’ll present a dead-simple, entirely actionable algo for timing the stock market.
So long as you choose an historically reliable valuation metric (see table below), it doesn’t make much difference which one you pick. For extra rigor you could assemble an index from several, rather than rely on a single statistic, but for the purposes of this dead-simple exercise we’ll use MADCAPE, an improved version of Robert Shiller’s CAPE.
MADCAPE can be calculated for the S&P back to the 1920s using freely available data, and it has a very strong inverse correlation (-0.9) with stock returns over the following 12 years. That means that if you knew MADCAPE at any time from 1928-2006, you could have predicted the annualized return of the S&P 500, including dividends, to within 3% over the next 12 years. That’s pretty darned accurate, considering that annualized real (after inflation) returns ranged from -2% to +15% over the set of possible 12-year periods (1929-1941 averaged -2% per year, while 1949-1961 and 1988-2000 returned +15% per year).
It’s clear that buying and holding an expensive market (like today’s) is a bad proposition, and that buying and holding a cheap one is a road to riches. This shouldn’t be so hard, but of course it requires utter conviction and scorn for the opinion of others. This is exactly what algos are made for, so let’s draw one up.
For simplicity’s sake, and because algos are rarely improved by complexity without giving up robustness, our algorithm will have just three conditions:
- Condition 1: Stocks are cheap, MADCAPE under 12
- Condition 2: Stocks are fairly valued, MADCAPE between 12 and 22
- Condition 3: Stocks are overvalued, MADCAPE over 22
Our model will check conditions once a year on December 31, and all adjustments will be made as of the first trading day in January.
- If stocks are cheap, we’ll put 80% of our money in the S&P (say with VTI or SPY) and 20% in 10-year Treasury notes (perhaps IEF or TLH).
- If stocks are fairly priced, we’ll keep 60% in the S&P and 40% in T-notes.
- If stocks are expensive, we’ll keep 20% in stocks and 80% in T-notes.
That’s the entire model. Let’s see how it has done over the last 87 years:
It outperformed the traditional static 60/40 portfolio, but experienced less downside volatility because it reduced exposure when stocks were expensive, but increased exposure when risks were lower.
Here’s a close-up of 1995-2016, which included three major bull markets and two major bear markets. Our algo just chugged along at its own pace, virtually unaffected by the dot-com and housing crashes:
In 2017, with the highest MADCAPE reading ever, the strategy is of course holding 80% in Treasuries, ready for another crash.
The model described above could be made more or less aggressive, as suits age or disposition, by adjusting the equity allocations or valuation thresholds. An aggressive model might go 100% into a cheap stock market, or even take a levered position, whereas a very conservative model might max out at 50%. Either way, the risk-adjusted return will exceed that of a comparable static allocation (for example, a static levered portfolio will get wiped out in bear markets, but by only using leverage in cheap markets an investor can reduce that risk).
I have no reservations about implementing a valuation timing model in real portfolios, though clients can get frustrated while lagging a runaway bull market. That is the price to pay for what sounds too good to be true: improved returns with less downside compared to a traditional portfolio that rebalances to static allocations. As with most such improvements, the real catch is emotional. The trick is to commit to the process and defer completely to the statistics. If you second-guess decisions or even start to think of them as your own, you’ll pay the price in anguish. It’s best to stop worrying and learn to love the algo.