Macroeconomic Dissociation

I’m lucky enough to have worked with some clients for many years, many through modeling relationships in which we have measured their key macroeconomic drivers.  We get to know these businesses in some depth during that time, and get pretty good at understanding the drivers.  But over the past six months we’ve seen changes in the key factors that surprised even our key analytic talent:  business performance is better than our math predicted. This effect, and even the timing of the effects were similar across two key clients in very different industries.

Let me be clear: we’re in the business of explaining performance, so when we get surprised, something is going on.

The usual suspects

In the past years, three typical macroeconomic factors that have influenced the broad swings in sales have been:

When those don’t completely account for business performance, we try others: gas prices, housing turnover, and any other available macroeconomic factors that could have a reasonable causal inference.  But of late we’re finding that none of our standard approaches fit- we’re still finding things like the below, in which actual sales consistently beat the model-predicted sales:

The chart above shows actual sales (blue line), the model-predicted sales (red line), and the model error (the bars at the bottom).   So when you see the bars high, that means that actual sales were higher than the model predicted.

Of course, we weren’t out of ammo and we worked with clients to find appropriate solutions to the model error.  But it’s also edifying to find that our sentiments are reflected by a CEO during his quarterly earnings conference call:

"We still look at this [macroeconomic indicator] to help understand
the pressures in the housing market and the fact that our
performance has disconnected from it suggests that our business
is stabilizing and can improve even as the [...] market
remains under stress."

What does it mean?

The bottom line is that consumers are acting (relative to the indicators, at least) in a way they haven’t in recent memory, pointing to the need for a new indicator that captures this changed behavior.