December 12, 2012
Stress-Testing Forecasts For Tomorrow's November Retail Sales Report
Will tomorrow's retail sales update for November bring pain or pleasure for the increasingly delicate and high-stakes art/science of deciding where the business cycle's headed? October's report was somewhat sobering, thanks to a 0.3% drop in consumption—the first since June. Some of the weakness was blamed on Hurricane Sandy. If so, will November's numbers bounce back after a month of relatively clear skies? Yes, according to the consensus forecast, which projects a handsome 0.4% rise, according to Briefing.com. Sounds good, but that's a bit high relative to a pair of econometric models I routinely use for additional context when considering where the data's headed.
I crunched the numbers in two ways, using vector autoregression (VAR) and autoregressive integrated moving-average (ARIMA) models, both of which also show up in my work with modeling the business cycle and nowcasting US GDP. As for today's retail sales projections, I'm using R software by way of the "forecast" and "vars" packages. These tools optimize the parameters based on the historical data records and so in some degree the resulting projections represent benchmarks for evaluating forecasts from other sources.
For the VAR projection of retail sales, I've chosen four data series to search for interdependent relationships: US private sector employment, personal consumption expenditures, disposable personal income, and the University of Michigan Consumer Sentiment Index. By contrast, the ARIMA projection uses only the historical data for retail sales as the basis for looking ahead. Here's how the two estimates compare for estimating the monthly change for November retail sales (in nominal dollar terms):
Averaging the two forecasts gives us a +0.1% estimate. Keep in mind that the confidence intervals for each of these point forecasts is, as usual, wide enough to keep us humble for thinking that the future is clear enough to put a high degree of confidence in one number. The potential for fairly large negative or positive surprises, econometrically speaking, remains substantial.
Nonetheless, my take on the these two forecasts is that we should be a bit more cautious in expecting a strong rise in tomorrow's report compared with the consensus outlook. November is likely to show a modest rebound from October, but I'm not expecting a large upside surprise.
Posted by jp at December 12, 2012 9:53 AM