Zero Hedge 
Wednesday, July 7, 2010
Goldman outdoes itself again. After Jan Hatzius has been banging the economic slowdown drums for days now, the firm’s other prominent economist Andrew Tilton is out with a new report “Recession Forecast Models Back in Vogue”, according to which the firm plugged in a few numbers into an overclocked iMac (appropriately equipped with the No Recession Ever ap), asked if there will be a recession, and the result was, stunning, “no way in hell.” Most hilariously, the report contains the following stunner: “Typical recession forecasting models estimate a near-zero likelihood that the economy has entered recession again, or that it will in the near future… The best news first: the model shows essentially zero probability that the economy is currently in recession. Payrolls have generally been expanding in recent months and the unemployment rate has actually come down slightly. This is unlikely to be a controversial conclusion for most market participants and so we will not dwell on it further.” In other words, because everyone knows that there is really no trouble in the jobless arena, aside from some rumblings in the periphery that the real unemployment rate is, oh, 16.5%, Goldman sees no need to discuss this data point, as it is really completely irrelevant. Oh yes, and the model refs out if you assume in negative input. Moody’s coupled with a dash of European stress tests anyone?
Andrew Tilton’s complete joke of a report below:
Last week’s thoroughly disappointing macro data set prompted many market participants to wonder anew about the risk of a “double dip” in coming months. One way to answer this question is via a quantitative “recession probability” model.
Typical recession forecasting models estimate a near-zero likelihood that the economy has entered recession again, or that it will in the near future. But they suffer from a serious bias: most models use the slope of the yield curve as a forecasting variable, with a flat or inverted curve a classic warning sign of a slowdown or recession. In the current environment of near-zero short-term rates, the yield curve must always have a positive slope, which normally implies a decent growth outlook and therefore leads such models to benign conclusions.
A forecasting model which leaves out the yield curve variable—reducing its accuracy under normal circumstances somewhat—and also adjusts for employment-related distortions currently estimates roughly a one-in-four probability that the economy will be in recession six months from now. However, we would not place too much weight on any one model, as it is difficult to capture idiosyncratic factors (such as the impending fade-out of fiscal stimulus or the risk of heightened concerns about sovereign debt sustainability) in a model estimated on the last 40 years of US data.
Last week’s macro data set was thoroughly disappointing, including worse-than-expected readings on the two most important monthly economic indicators: the ISM manufacturing index and the employment report. The ISM index dropped 3½ points and featured a seven-point decline in the “new orders” component, which has the most predictive power for future industrial activity. As for the employment report, private-sector payrolls grew by only 83,000—less than needed to keep up with population growth over the long term—and the workweek shortened. The unemployment rate declined to 9.5%, but only because a large number of people gave up looking actively for work; the overall employment-to-population ratio has moved halfway back to its late-2009 lows.
These disappointments have prompted many market participants to ask anew about the possibility of a “double dip” in the economy. We interpret this below as the probability that the economy re-enters recession, roughly defined as two or more consecutive quarters of contracting real GDP. (Conversations with clients suggest that some refer to any significant renewed slowdown as a “double dip,” even if it does not qualify as a full-blown recession.) To provide some perspective on this question, we have dusted off a recession forecasting model that we employed in 2007-2008 (see “Recession Forecasting Models Flashing Yellow,” US Economics Analyst 07/39, September 28, 2007).
(ARTICLE CONTINUES BELOW)
Our recession forecasting model contains the following variables: the slope of the yield curve (10-year vs 3-month Treasury yields), the recent change in the unemployment rate, nonfarm payroll growth, the recent change in equity prices, the year-over-year change in housing starts, and the ISM manufacturing index. Generally, the labor market variables tend to be more significant as the forecasting horizon shortens, while the slope of the yield curve is the best predictor of a recession six to twelve months out. Although versions of the model were calibrated to forecast recessions at different horizons, we focus here on a) the probability that the economy is currently in recession, and b) the probability that the economy will be in recession six months from now. We estimate the model over the period from 1970 through mid-2009 using real-time rather than revised data (some adjusted versions, discussed below, begin in the late 1970s due to data availability).
The best news first: the model shows essentially zero probability that the economy is currently in recession. Payrolls have generally been expanding in recent months and the unemployment rate has actually come down slightly. This is unlikely to be a controversial conclusion for most market participants and so we will not dwell on it further.
More surprisingly, the model also shows a very low probability (1.6%) that the economy will be in recession six months from now. Why so low? The aforementioned labor market variables are one reason, but another important one is the positively sloped yield curve. As the yield curve embeds expectations of future Fed policy, a flat or negatively sloped curve is a sign that the market sees Fed easing as likely—a classic sign of a slowing economy.
The problem, of course, is that the yield curve must be positively sloped in an environment of near-zero short-term interest rates. Even if the market views the growth outlook as terrible, the Fed cannot cut rates, so risks to future short-term rates are one-sided. In normal times, the current positively-sloped yield curve would imply a decent growth outlook, but this is clearly misleading. It would be nice to find another yield-related variable without this bias, but none we have tried are nearly as significant. Still, we can gain a sense of how important the bias may be—at the price of a moderate loss in accuracy under normal circumstances—by simply removing this variable from the model. Doing so raises the six-month recession probability to 8.6%.
The employment-related variables in the model also suffer from some distortions at present. Because of short-term Census hiring, nonfarm employment posted large gains through May and declined in June. Private-sector payroll growth provides a less noisy barometer of underlying job market trends at the moment. Another issue concerns the unemployment rate; because a large number of people have left the labor force (more than a million over the past year), it may not reflect the full extent of labor market weakness. An alternative is the employment-to-population ratio. Substituting these variables in our original model (with the yield curve) gives a recession probability of 6.1%. So far, not so bad, but if we combine the yield curve adjustment and the employment-related adjustments, the model generates a 23.9% probability that the US economy will be in recession in six months’ time. In other words, it suggests recession is not a base case, but is clearly more than just a tail risk.
Lest any reader take these forecasts too literally, we hasten to emphasize that any quantitative model can only capture a subset of the potential factors that could push the economy into recession—or keep it growing at a healthy pace. The substantial changes in probability just from the changes discussed above illustrate the potential instability in models and the need for robust quantitative and qualitative analysis. Our subjective assessment of recession probability leans to the high side of the model results. This is partly because the model focuses on a fixed six-month horizon; the probability will naturally be higher over longer horizons. In the current episode, the fade-out of fiscal stimulus, the wind-down of the inventory cycle, and the risk that sovereign default concerns further tighten financial conditions are all reasons for concern—but these are difficult to include in a quantitative model that covers only the last 40 years of US economic data.