Inflation modelling

As a quick and dirty attempt, using the Alpha Vantage API, I was able to source some US Data: 1.FED rate, 2.unemployment, 3.CPI, 4.payroll and 5.retail sales. I turned the non-stationary data (CPI, payroll, retail sales) into 12 month growth. Then I made the prediction target the forward CPI growth for the next 12 months. I tried linear regression and XGBoost.

The best model was XGBoost, the model degrades (Rsquared) when looking at the development vs out of sample (data it has not seen before). But as a first attempt it's not terrible. The model would benefit from more features, for example the popular "Personal Consumption Expenditures" favoured by the FED over CPI because it factors in basket substitution. US headline and Core inflation as of Oct23 has shown reasonable abatement since its recent peak.  The model's prediction of the CPI reaching 2% levels by next year is not terribly implausible, this is also close to a forecast by the Cleveland FED. https://www.clevelandfed.org/indicators-and-data/inflation-expectations

Obviously this is all guess work and forecasts are well known to frankly suck. There are significant risks of protectionism/supply chain shocks/etc. and these risks should make investors more cautious and demand a commensurate risk premium.