Backward looking forecasts

We have long written about the dangers of economic forecasting as an input into the investment process. The future is often highly surprising and nobody has a perfect model of how the economy works, which means forecasts will usually turn out to be wrong  – just ask any central banker.

The below chart from the Reserve Bank of Australia shows just how difficult they have found it to forecast domestic wages. Any follower of the Bank of England’s forecasts or the Federal Reserve’s ‘dot plots’ can relate to this.

Have another go

From an investing standpoint, the problem is twofold. Not only is any one economic forecaster unlikely to have a systematic ability to out-predict the competition, but even if they did, asset prices behave in surprising ways too.

Knowing that you correctly predicted a Trump victory provides cold comfort if it was the basis of a decision to sell equities in advance of last November’s election. Similarly, deciding to buy the US dollar in December on a prediction that the Fed would raise rates more than expected over the next 3 months hasn’t worked, so far at least.

Not only is predicting the future ridiculously challenging, but working out what happened in the recent past is difficult enough. Admirably, various Federal Reserve banks now provide ‘nowcasts’ of US GDP growth using statistical formulae which are updated with each individual economic data release to provide a live estimate of growth over the current quarter.

What is interesting is that these Reserve Banks, staffed by some of the best minds within the economics field, can have competing models which produce very different estimates of what has happened to the US economy over the past 3 months. Even though we are at the end of March, there is stark disagreement about what US growth was in Q1.

We can’t even forecast the present

Currently, the Atlanta Fed’s GDP Now series is estimating below-trend annualised growth in Q1 of 1%, while the New York Fed’s Nowcast estimates a very encouraging 3% growth rate. This is a huge difference given that both models are likely to be using very similar inputs.

That two sophisticated models can provide such divergent estimates of backward looking data just goes to show how hard it is to estimate the aggregate economy, let alone forecast its future direction.