Anomalies abound in today’s economy. Can artificial intelligence know what’s going on?

All the fuss today is about machine learning and ChatGPT. The algorithms associated with them work well if the future is similar to the past. But what if we are at an inflection point in economic and political conditions and the future is different from the past? Will record profit margins, inflated asset prices and low inflation and interest rates of the past 30 years be an accurate reflection of the future? Is this time different?

Maybe we’re already there. Things do not seem to make sense anymore. Have you noticed that economic indicators seem to have stopped working as well and as predictably as they have in the past?

Here are some examples of the puzzling behaviour of economic statistics of recent months.

An inverted yield curve has historically been a good indicator of recessions. For several months now the yield curve has been inverted and yet the U.S. economy has been adding millions of jobs, leading to an historic low unemployment rate. Employment is booming while the economy at large is not.

Consumer sentiment, as reflected in the University of Michigan surveys, and consumer spending have tended historically to move together. But this time around, while consumer sentiment took a nosedive, consumer spending and credit card balances keep growing, reaching record highs.

Construction employment and homebuilder stocks are rising while housing permits and housing starts are falling. Normally, homebuilder stock prices would reflect the collective wisdom of financial markets about housing activity. Not this time.

Bond markets are expecting inflation to recede to the Fed’s target rate of 2 per cent. In this case, the real interest rate, implicit in the 10-year treasuries yield of between 3.5-4 per cent, is 1.5-2 per cent, which is close to historical averages. But prior to the Silicon Valley Bank debacle, some surveys pegged expected inflation to about 3 per cent going forward. Assuming the real rate is the same, this implied a 10-year treasuries yield of between 4.5-5 per cent. Either the bond market was out of line or forecasters’ inflation models do not work as well as in the past.

And oil prices are around US$70 a barrel despite the recent banking crisis and at a time when the economy is slowing down and believed to be entering a recession. Based on past experience at this point in the business cycle oil prices should be at US$50 or less. But they are not. Which begs the question: What will happen to oil prices when the economy enters a growth phase, especially with the opening of China after the COVID-19 lockups?

And the list of puzzling contradictions goes on. Having said that, someone may argue that the labour statistics, for example, are a lagging indicator and show where the economy was, not where it is going. While this is true, the magnitude of divergence between labour statistics and economic activity is so much higher than they’ve been historically. That makes one wonder what is going on.

It could be that many of these puzzling statistics are the result of “survey fatigue,” as Bloomberg Businessweek calls it. The publication reports that there has been a decline in response rates for many surveys government agencies use to collect economic data.

For example, employer response to the Current Employment Statistics survey, according to the publication, which collects payroll and wage data each month, has declined to under 45 per cent by September, 2022, from about 60 per cent at the end of 2019. The issue here is the non-response bias: that people who are not responding to the survey are systematically different from those who do, and this skews results. Could weakening trust in institutions and governments be behind the decline in response rates in recent years? If this is the case, the problem is serious and difficult to reverse or eliminate.

As a result, machine learning algorithms that need massive and good quality data about the past and assume that the future will look pretty much like the past may not work. Then what? Should we re-examine our old models? Or will human intervention always be required? Machine learning will not be able to replace investor insight and “between the lines” reading of nuanced economic numbers.

George Athanassakos is a professor of finance and holds the Ben Graham Chair in Value Investing at the Ivey Business School, University of Western Ontario.

Source link