Bridge over ocean
1 April 2016 CFA Institute Journal Review

Occam’s Razor Redux: Establishing Reasonable Expectations for Financial Market Returns (Digest Summary)

  1. Antony Jackson, CFA

In a series of articles in the early 1990s, John Bogle presented two methods for forecasting long-run stock and bond returns. Two decades’ worth of out-of-sample data validates the virtues of the simplicity and the forward-looking nature of the original models. The low prospective returns on mixed stock/bond portfolios are likely to refocus investors’ attention on the costs of mutual fund investments.

What’s Inside?

The authors revisit two simple models for forecasting stock and bond returns over a 10-year horizon. In the case of the stock market, future performance is a function of the current dividend yield, prospective earnings growth, and the likely level of the future P/E. In the bond market, the key indicator is the initial yield.

How Is This Research Useful to Practitioners?

Anyone who manages a defined benefit pension plan (or who is relying on the proceeds of a defined contribution plan in their retirement) will be concerned by the authors’ forecasts for the coming decade. They estimate that an investment portfolio consisting of 50% stocks and 50% bonds will only produce an annual return of 4.5%. Taking into account management fees, portfolio turnover costs, and other charges, this low return means investors face losing around 50% of their annual returns to costs.

The authors’ model for stock returns is the sum of the current dividend yield, the expected annual growth in nominal earnings per share, and the expected change in the price-to-earnings multiple. The model bears some similarity to the Grinold and Kroner (Investment Insights 2002) model, but without the “repurchase yield” as a result of stock repurchases.

The model for bonds is even simpler; it is the current yield on the long-term Treasury bond. Deviations of realized performance from forecast performance are essentially attributable to reinvestment risk.

How Did the Authors Conduct This Research?

The authors test the models’ performance by considering rolling 10-year windows, both for the two decades’ worth of data since the original articles were written (1990) and for a longer backtesting period of 1906 to 1989.

The stock market model is based on the S&P 500 Index. The forward-looking parameters are the initial dividend yield, the average 10-year trailing earnings growth, and an assumption that the forward P/E reverts to its 30-year average. For the 1990–2014 period, the model predicts annual stock market returns of 9.2% versus realized returns of 9.6%. Interestingly, the most challenging decade for the market was immediately after the model was published because the P/E persisted well above its long-run average. For the overall 1915–2014 period of realized returns, a regression of actual returns on predicted returns has a correlation of 0.67 and an R2 of 0.44.

The bond market model focuses on the 10-year Treasury bond. There is an exceptionally high correlation of 0.95 and an R2 of 0.90 for the regression of the 10-year Treasury return on the initial yield.

Abstractor’s Viewpoint

The authors argue a compelling cost-based case for passive index investment, attributable in part to lower management fees, but also to much lower portfolio turnover costs. This view is perhaps a little unkind to the role of active management—for example, consider the recent regulatory concerns with overcrowding and herding in the ETF space. I would also like to see the model pitted against some alternatives, such as Robert Shiller’s cyclically adjusted P/E or the Grinold and Kroner model. Overall, the authors succeed admirably in demonstrating the power of Occam’s razor, which is the reductionist notion that explanatory assumptions should be kept to a bare minimum, in US stock and bond markets.

We’re using cookies, but you can turn them off in Privacy Settings.  Otherwise, you are agreeing to our use of cookies.  Accepting cookies does not mean that we are collecting personal data. Learn more in our Privacy Policy.