Nowcasting Mexico's Quarterly GDP — Banco de México's Bridge-Equation Model Beats Bloomberg and INEGI's Rapid Estimate

Mexico's central bank needs GDP estimates well before INEGI's official quarterly release, since monetary-policy decisions cannot wait for the standard multi-week publication lag. Economist Oscar Gálvez-Soriano, working within Banco de México's own economic research directorate, built and tested a set of real-time GDP-nowcasting models using monthly indicators such as industrial production, retail sales, employment, trade and financial data.
Five nowcasting approaches were compared over the full 1993 Q1-2017 Q1 sample: one dynamic factor model (DFM), two bridge-equation (BE) models and two principal-component-analysis (PCA) models, benchmarked against a simple AR(1) model, using three information sets of 25, 8 and 11 monthly indicators respectively.
The average of the two BE models produced the lowest mean squared error overall (MSE = 0.026) and was statistically more accurate than the DFM and PCA models under the HLN-modified Diebold-Mariano test. In a genuine real-time test spanning 12 consecutive quarters (2014 Q2-2017 Q1), the BE average's MSE of 0.004 beat Bloomberg's analyst survey (0.019), Banco de México's own Survey of Professional Forecasters (0.051) and INEGI's rapid GDP estimate (0.015); three-quarters of the BE forecasts landed within 0.1 percentage points of the eventual official GDP figure a full month before its release.
The authors caution that these results rest on deliberately small, hand-picked indicator sets — chosen partly to avoid the overfitting that hurt larger factor models — and have not been shown to generalise to other developing economies with sparser monthly data; the study, published as a peer-reviewed research paper, does not itself confirm whether the bridge-equation approach has since become Banco de México's standing operational nowcasting tool.

Read the full analysis: https://www.scielo.org.mx/scielo.php?script=sci_arttext&pid=S0186-72022020000200213