Recession United States on the radar

While our warnings for a US recession were initially met with skepticism, we are increasingly getting the question whether the US economy is not already in a recession. Recent data have been disturbing. Nonfarm payroll growth slowed down to 20K in February from 311K in January. Retail sales fell by 1.6% in December, followed by only a modest 0.2% rebound in January. In fact, the Atlanta Fed’s ‘GDPNow’ nowcast for Q1 stood at only 0.4% on March 13. In this special we explain why we think that we are heading for a recession in 2020, but at the same time we doubt that the recession has already hit the US.

The reason why we think that the US will fall into recession in 2020 is the shape of the yield curve. In general, economists are not very good at forecasting recessions. Large macroeconometric models – and consequently economic institutions – rarely forecast a recession, unless it is obvious to everybody that a recession is near or has already started. Usually, these models are mean-reverting and the impact of economic shocks tend to fade. Consequently, they are not good at spotting turning points in the economy. However, there is substantial empirical evidence in the economic literature that inversions of the US treasury yield curve are good predictors of US recessions. That is why we built a recession forecasting model based on the yield curve a few years ago. At the start of this year the recession probability generated by our model reached 69%. Hence we predicted that the US economy would be in recession in 2020. We strengthened our case pointing at the first cracks in the housing market in 2018.

Despite the robust empirical evidence that yield curve inversions predict recessions, the Fed is still thinking that ‘this time is different.’ According to various Fed speakers QE has reduced term premia and therefore we should not interpret yield curve inversions as harbingers of a recession. Ironically, the Fed told us the same thing before the Great Recession. In 2006, Fed Chairman Bernanke said that ‘the global savings glut’ was depressing term premia and therefore ‘I would not interpret the currently very flat yield curve as indicating a significant economic slowdown to come’ (March 20, 2006). Interestingly, at a July 2018 event to discuss the lessons from the 2008 financial crisis then ex-Chairman Bernanke said that ‘historically the inversion of the yield curve has been a good sign of economic downturns, this time it may not.’ While the flattening of the yield curve could not convince the Fed, the stock market did. After steep declines in the S&P500 in December 2018, the Fed decided to pause its hiking cycle. Ironically, the stock market has a worse record of predicting recessions than the yield curve.

The optimal forecasting horizon of our yield curve based recession probability model is 17 months. This means that it provides an early warning signal. The drawback is that the model does not tell us very much about much shorter horizons, or about the present. Does this mean that after receiving the early warning signal we have to wait for 17 months to see if the recession materializes? When we asked ourselves this question a few years ago, we decided to expand our toolset with another forecasting model that can be based on the economic literature. There is empirical evidence that the spread between US commercial paper and treasury bills has predictive power for recessions at shorter horizons. Therefore, we tried to predict NBER-recessions 1 to 5 months ahead with the 3 month paper-bill spread. We found that the optimal forecasting horizon of this financial instrument was 3-4 months, which seems plausible given its 3 month maturity.

To complete our recession warning system we added the nowcasting model we developed in early 2008. At the time, the US economy had deteriorated severely and we needed a tool to identify when we would be in recession. After all, if we went by the GDP data there would be at least a 6 month lag between the start of the recession and the recession showing up in the data as two subsequent quarters of negative growth. And the NBER usually takes even longer (6-18 months) to date the start of a recession. To be able to determine the recession in real time, we estimated a model that uses a number of monthly variables to explain NBER-dated recessions. Our model concluded that a recession had started in February 2008. Much later, in December 2008, the NBER officially chose January 2008 as the first month of the recession. A few years ago we re-estimated this nowcast model to complete our recession forecasting system.

By combining the yield curve based recession probability model (13-17 months ahead), the paper-bill based model (1-5 months ahead) and the nowcast model (0 months ahead), and interpolating the horizons in between we built a rudimentary ‘recession radar’ that provides a term structure of recession probabilities. While the recession probability peaks in the summer of 2020, it is still low for the first half of 2019. At present, our nowcast model gives only a 4% probability that the economy is in recession. This is only slightly higher than the 3% probability at the 3 month horizon implied by the paper-bill spread. Looking further ahead, our model indicates a 60% probability of a recession by July 2020.

The weakness in recent US data, confirmed by the Atlanta Fed nowcast of 0.4%, is also likely to be temporary because of seasonality and the recent government shutdown. As we have shown repeatedly, there is residual seasonality in the GDP data published by the BEA with Q1 growth systematically lower than other quarters. This is due to imperfect seasonal adjustment in the underlying data that are used to calculate GDP. Therefore, we should expect Q1 growth to be lower than other quarters, and Q2 to be higher than Q1 on average. What’s more, the government shutdown is estimated to have subtracted 0.4 percentage points from Q1 GDP growth (according to the Congressional Budget Office), but it is also expected to add a full percentage point to GDP growth in Q2. Consequently, both seasonality and the government shutdown partially explain Q1 weakness and suggest a rebound in Q2.

While our recession radar is still rather rudimentary, with various horizons interpolated instead of estimated, for now it confirms our view that we are headed for a recession in 2020 and that the current weakness is not yet the start of the recession, but rather the first cracks in the long economic expansion (by July 2019 the longest since WWII) that followed the Great Recession. Going forward, it may be useful to further develop our recession radar. Moreover, we will take a closer look at various channels through which the recession could be amplified. For now, we assume a run-of-the mill recession in 2020H2 caused by a monetary policy mistake. However, there are several potential bubbles – such as in leveraged loans and in commercial real estate – that could deepen the recession through a financial crisis.