Ai accelerator revolutionises econometrics

How to overcome the limitations of traditional econometric models that underlie the most relevant macroeconomic variables, from GDP to inflation, from unemployment to retail sales?

A concrete proposal comes from a study presented at Luiss, during a conference entitled 'The AI accelerator, how to integrate econometric forecasts', in which the contents of a project jointly developed by the IIEC (Italian International Economic Center) and the Luiss Quantum & AI Lab, directed by Antonio Simeone, within the Luiss AI4Society Research Centre were illustrated.

A project that aims to integrate classical econometrics, to maintain the explanatory value of the relationships between variables, by including alternative data such as ship and air traffic, energy consumption, online sales, short selling, collected with 'high granularity', advanced Artificial Intelligence, ecophysics, used in particular to analyse trade flows and systemic resilience, through indicators based on entropy.

The first results of the AI Accelerator are surprising. "The new model," observes Giuseppe Italiano, pro-rector for Artificial Intelligence at Luiss, "has demonstrated a predictive capacity for Italian GDP that is superior to that of traditional benchmarks, especially in moments of economic shock such as the pandemic crisis.

The model would have predicted in advance a GDP contraction of 7 per cent, which actually turned out to be 9 per cent. "We are extending the model to forecast inflation, and will make it usable through an operational dashboard, to give concrete tools to companies, funds and public decision-makers. The problem,' Simeone notes, 'is that traditional econometric models use official data that are often 'lagged' and linear approaches. We have seen how, for example, central banks have made mistakes in predicting inflation.