How To Use Shrinkage To Fine-Tune Your Forecasts

By Tim Kelley | More Articles by Tim Kelley

The acclaimed economist, John Kenneth Galbraith, once quipped: “The only function of economic forecasting is to make astrology look respectable.” Sometimes, it feels like the same can be said of investment forecasting, because predicting the future is not easy.  But there are tools we can use to help minimise errors.

Good forecasting is one of the critical foundations of value investing.  In most cases, an informed view on valuation needs to be built on an appropriately rigorous valuation model, whose outputs are likely to be highly sensitive to the assumptions chosen – assumptions like revenue growth rates, and earnings margins. It follows that being able to accurately forecast these sorts of parameters is central to getting to a sound investment decision.

These forecasts usually need to reach well into the future, and, of course, the future is not ours to know.  For example, while an industry might have enjoyed a period of “comfortable” competition for some years, a gradually changing industry structure may be leading to the emergence of new and credible competitors who might price aggressively to gain share. Alternatively, we might see that one competitor is working to take costs out of their business, which could lead to materially higher margins for that company. In both cases, we can see clearly what is there today; we can only glimpse what may unfold in future.

When dealing with uncertainty like this, one helpful concept to keep in mind is something called “shrinkage”.  Shrinkage effectively asks you to explicitly adjust your forecast to take account of the level of uncertainty you see.  For example, if we see that company A has a profit margin of 20%, but we calculate that a cost reduction exercise could increase margins to 30%, our initial view might be to project margins rising to 30% over a period of time.

Shrinkage asks us to consider how much confidence we have in our ability to forecast that change in margins, and (and this is the important bit) to adjust our forecasts to account for that.

If we had perfect foresight and were certain that the world would unfold according to our calculations, we could be justified in forecasting 30% margins.  However, what should we do if we don’t have perfect foresight? At the other end of the spectrum, we might feel that our ability to forecast a change in margins is effectively zero.  In that case, the correct approach might be to forecast no change to the existing margin of 20%.

In most cases, we fall somewhere between these two extremes, and the correct approach is to weight our forecast according to the level of confidence we have, so in this case something closer to 25% is probably the way to go to minimise forecast (and investment) errors.

Acknowledging the limits of our forecasting ability can be uncomfortable, but the underlying principle has some undeniable logic: set a baseline on what you know to be true (eg current margins), and be a little conservative in deviating from that baseline in response to things that are uncertain.  It is all too easy to frame forecasts as if we have perfect insight, and the result can be valuations that fall far above, or below, a reasonable estimate of “fair value”.