Crockpot: History is all we have to go by

Many investment planners still use long term historical returns to estimate future market returns on the basis that, if the historical period is long enough, everything smoothes out. This is pure bunk.

In order to demonstrate this as conclusively as possible, we illustrate the impact of using the 20-year historical return of six major equity markets as forecasts for their return for the subsequent decade, over the past 107 years.

Figure 1 shows the average real returns for the 20 years prior to a reference date. This return becomes the forecast for the next ten years. It is then compared to the actual returns in the ten years after the reference date.

Needless to say, the results are horrible! The shaded boxes highlight where the forecast error is both big and in the wrong direction - that is, either the forecast (based on the past return) was for below average returns but the actual return was high, or vice versa. It's clear that the forecasts were badly wrong, most of the time.

And look at the size of the errors! On average, had you used the 20-year historical return as the forecast for the next decade, you'd have been out by over 6% per annum in Australia, over 8% per annum for US equities, and more than 10% per annum off for Japanese and German equities. While the latter two markets were clearly affected around the war years, using just the last 30 years would have given average errors exceeding 10% per annum in each market.

Finally, if you're hoping that diversification would reduce the size of the error, think again. On average, the forecasting errors for World Equity returns average about 8% per annum as well.

So, using long-term historical returns as a forecast, you are, on average, going to be wrong by almost 8% per annum for ten years. A forecast of 10% per annum is likely to end up at 2% per annum (or worse). A forecast of 3% per annum real returns is likely to be 11% per annum or more. How can anyone make sensible plans with such horrible forecasts?

A simplistic mean reverting alternative is to take the past 20 year real return, and add back the average rate of over or underperformance compared to 6% per annum - the size of the forecast errors will almost halve. And, using the methods such as farrelly's employs, we can do quite a bit better than that.

Using past returns – even 20-year past returns – to forecast the future? It's nuts and you can clearly see it's nuts!

Figure 1: Using the prior 20-year return to forecast the next 10-year real return (%)

	Australia		Germany		Japan		US		UK		World
	fcast	actual	fcast	actual	fcast	actual	fcast	actual	fcast	actual	fcast	actual
1920	7.8	16.3	-4.9	5.8	9.4	2.5	2.5	14.4	0.2	9.3	0.7	13.0
1930	9.9	9.5	-3.9	6.5	4.7	10.4	5.6	1.9	3.9	2.6	4.3	1.7
1940	12.8	3.2	6.1	-10.3	6.4	-25.7	8.0	4.0	5.9	3.1	7.2	1.0
1950	6.3	8.9	-2.3	24.6	-9.4	27.5	2.9	15.7	2.8	13.7	1.3	17.1
1960	6.0	10.6	5.7	3.9	-2.7	8.5	9.7	5.6	8.3	6.5	8.8	5.5
1970	9.7	-4.6	13.8	-2.5	17.6	3.4	10.5	-0.7	10.0	-1.4	11.1	0.9
1980	2.7	8.6	0.6	14.0	5.9	18.2	2.4	11.0	2.5	15.4	3.2	13.5
1990	1.8	8.1	5.4	11.8	10.6	-7.1	5.0	11.4	6.7	9.3	7.0	0.4
2000*	8.3	12.4	12.9	1.6	4.8	5.9	11.2	0.9	12.3	2.6	6.7	2.8
Av error		6.3		12.6		15.3		8.1		6.9		7.9

[Source: E. Dimson, Marsh, P. and Staunton M., 2002, Triumph of the Optimists: 101 Years of Global Investment Returns, Princeton University Press.]

Notes
The Forecast is the average real return per annum for the 20 years prior to the reference date.
The Actual is the average real return per annum for the 10 years after the reference date.
* Period is December 2000 to April 2007

Tim Farrelly is principal of farrelly's, the Australasian financial services industry's first dedicated asset allocation research house.

Shared membership details

Read the paper