Can a Transformer “Learn” Economic Relationships?

This post is joint with Alex Imas — subscribe to his Substack here. In the 1960s and 70s, the standard way to do economic analysis was to feed data into a model and then estimate the relevant parameters. For example, based on historical data, economists believed there was a stable tradeoff between unemployment and inflation, i.e., the Phillips curve. Policymakers believed they could exploit the tradeoff by accepting higher inflation in exchange for lower unemployment. However, when oil price shocks in the 1970s generated both inflation as well as rising unemployment, expansionary monetary policy failed to deliver the predicted results. Instead of lower unemployment, the economy experienced “stagflation” of higher inflation combined with high unemployment, suggesting that the historical relationship had broken down. So what happened? Expectations of inflation made workers and firms change their behavior: knowing that inflation was going to be higher in the future, workers demanded high

This post is joint with Alex Imas — subscribe to his Substack here. In the 1960s and 70s, the standard way to do economic analysis was to feed data into a model and then estimate the relevant parameters. For example, based on historical data, economists believed there was a stable tradeoff between unemployment and inflation, i.e., the Phillips curve. Policymakers believed they could exploit the tradeoff by accepting higher inflation in exchange for lower unemployment. However, when oil price shocks in the 1970s generated both inflation as well as rising unemployment, expansionary monetary poli