Paper: A controlled linearized kalman filter for economic forecasting and adaptive modelling

  • Authors:

  • L. F. Pau

  • Affiliations:

  • Massachusetts Institute of Technology (Room 35-314), Cambridge, Massachusetts 02139, USA and E.N.S.Télécommunications, 46, rue Barrault, F 75013 ParisFrance

  • Venue:
  • Automatica (Journal of IFAC)
  • Year:
  • 1978

Quantified Score

Hi-index 22.14

Visualization

Abstract

An adaptive controlled forecasting procedure is described, which is specifically tailored to some major restrictions on economic data series and the resulting modelling tasks: (a) the periodically changing content of the economic statistics delivered and changing model structures, are represented by periodically changing transition and measurement equations; (b) jumps in the instrumental variables, decisions and exogenous quantities, are being compensated for by a model-based sensitivity correction; (c) quick model adaptation achieved via recursive estimation (dual Kalman filtering) of an autoregressive relation between all past recent measurements. This forecasting model presents therefore the original feature of being a mixed model-based-and-autoregressive (naive) prediction scheme. Considering a truly non-linear macroeconomic model, and a 20% jump on all instrumental variables, good forecasting performances are demonstrated numerically.