A0437
Title: A neural Phillips curve
Authors: Philippe Goulet Coulombe - Université du Québec à Montréal (Canada) [presenting]
Abstract: Many problems plague the estimation of the New Keynesian Phillips curve. Amongst them is the hurdle that the two key components, inflation expectations and the output gap, are both unobserved. Traditional remedies include creating reasonable proxies for the notable absentees or extracting them via some form of assumptions-heavy filtering procedure. Essentially, this is all unsupervised learning. We move towards a supervised extraction of inflation key drivers by developing a Hemisphere Neural Network whose peculiar structure allows the interpretation of the last layer's cells output as key macroeconomic latent states. Many benefits come from this neural approach. First, nonlinearities are trivially allowed for. Second, computations are quick and done within standard deep learning software. Lastly, the model typically forecast better than a wide array of benchmarks (including plain neural nets) while being interpretable.
