Wolf! Wolf! Wolf? Increasing US recessionary risks

9 May 2018 19:04 RaboResearch

In this publication we look at three different indications pointing at increasing US recessionary risks. Our treasury yield curve model suggests a 27% probability of a recession in the 17-month window. This probability is much higher than the recession expectation of the NY Fed.

G. Washington on a  1 dollar bill combined with Corona / Covid-19 newspaper headlines and a red declining stock market trendline

The economist who cried recession

The fable of the boy who cried wolf appears to be as old as approximately 500BC and has been rehashed in English poetry and even as a Looney Tunes classic cartoon. For those who are not familiar with the story: it’s about a shepherd boy who falsely warns his fellow villagers that wolves are attacking the flock. When the wolf does show up and the boy once again calls for help, nobody believes him anymore and both the sheep and the boy are eaten by the wolf. The moral of the story is that a liar will not be believed, even when telling the truth. It is for this reason that economists are very reluctant to cry recession. Indeed, economists have a bad track record in predicting both the timing and the magnitude of downturns. In fact, it usually takes them several years and numerous journal articles to explain why the recession occurred in the first place. To be clear: we have not penciled in a US recession in our forecasts. In this Special, however, we ask ourselves the question: what are the odds of a recession occurring in the near future?

Approaching the inflection point?

Many leading economic organizations are positive about the growth trajectory of the US. In April’s World Economic Outlook, the IMF is expecting US GDP growth to pick up to 2.9% this year and 2.7% next year. In March, the OECD also revised its forecasts upwards to 2.9% in 2018 and 2.8% in 2019. And the most positive of them is the Congressional Budget Office, which forecasts GDP growth of 3.0% in 2018 and 2.9% in 2019. In light of all this, our own projections (2.7% in 2018 and 2.3% in 2019) look kind of bleak, but nevertheless are still quite positive. At the same time, there are several indications that recessionary risks might be on the rise. Below, we separately look at three aspects: delinquencies on consumer loans, financial market indicators and the labour market.

Delinquency on consumer loans

Although most media attention was given to the 10-year US Treasury yield, which broke through the psychologically important 3%-barrier in April, short-term interest rates have been rising at an even faster pace. The expectations of further Fed tightening for the remainder of this year rose to a record high of 57 basis points last week, which implies that the market is increasingly factoring in a total of four rates hikes for this year (see figure 1). We don’t see the fourth hike in December eventually materialize, but rising short-term rates and a steepening in the front-end of the curve mean that consumer borrowing costs are already edging higher. This is exacerbated by the rising spread of 3-month Libor versus the equivalent overnight index swap rate, currently at 50 bps.

Figure 1: Fed fund futures are currently pricing another 57 bps of hikes this year

Source: Bloomberg, Macrobond, Rabobank

Although we don’t see a serious pickup of delinquency rates in various classes of consumer credit, e.g. mortgages and credit cards, delinquency rates on car loans have been on an upward trajectory ever since the Fed has started its monetary tightening cycle last year (figure 2). Moreover, in absolute terms, both credit card debt and serious delinquencies on car loans are already moving in global financial crisis (GFC) territory (figure 3). This means that households are increasingly taking on more credit card debt, but at the same time have difficulties servicing part of their debt (in this case: car loans).

Figure 2: Delinquencies of 90 days or more on consumer loans

Source: Federal Reserve Bank, Macrobond, Rabobank. Note: the delinquency rate is defined as the number of loans with delinquencies (of more than 90 days in this graph) as a percentage of total loans.

Figure 3: Households are taking on more credit card debt, but at the same time have difficulties servicing car loans

Source: Federal Reserve Bank, Macrobond, Rabobank

What’s more, household debt currently stands at 87% of GDP, which is broadly equal to the levels seen in 2006. An argument often heard is that these debt levels are covered by higher liquid asset values (figure 4). This might also explain the low household savings rate of barely 3%. Households feel less need to save a substantial part of their income when the value of equities and other financial entitlements are pushing higher. These net worth statistics, however, mask substantial underlying inequality. The 1% richest group of households controls approximately 40% of wealth (see Federal Reserve, 2017 and – we didn’t make this up! – Wolff, 2017), and according to the latter study the top 20% even holds 90% of all wealth. Especially for the lower income groups, low savings rates today might turn out problematic if interest rates keep rising in the future.

Figure 4: Low savings rate and elevated assets move in opposite directions

Source: BIS, Macrobond, Rabobank. Household liquid assets encompass deposits, debt securities, corporate and non-corporate equities, mutual fund shares and miscellaneous.

Financial market indications

The current developments also imply that the Treasury yield curve has continued to flatten. Historically, inverted curves have been one of the better predictors of recessions, which is the reason why this curve is being watched closely, also by the Fed. In fact, each US recession since the 1960s was preceded by a yield curve inversion. With around 50 basis points to go, it will still take some time before the 2s10s actually inverts. However, the flatness of the yield curve does indicate an elevated risk of recession. Can we attach a numerical value to this risk?

In an earlier study we found that the spread between the 12 month and 10 year US treasury yield has been the most reliable predictor of US recessions in comparison with other spreads. Moreover, the optimal horizon at which the slope of the yield curve is able to predict recessions is 17 months, well beyond the 12 month horizon that prevails in the literature. If we take the values of the yield curve by the end of April, the optimal horizon is September next year. By estimating a logit model it is possible to translate the current 12m-10y yield spread into a recession probability. Our own US recession forecasting model (see figure 5), puts the probability of a US recession at 27% by September 2019. As long as the recession probability remains below 50% the recession forecasting model supports Rabobank’s baseline forecast of continued economic expansion. However, at 27% it also indicates that there is considerable downside risk to the baseline. We are also more bearish in our recessionary expectations than the NY Fed, which puts the probability of a recession happening in 12 months at roughly 11%.

There are also early warning signs of an inversion in the OIS forward market. Although the one-month OIS forward curve is a constructed curve that is hardly actually traded, we could see the inversion that is visible in the 2-6 year domain as a signal that the market starts to think about rate cuts after two years (figure 6). There weren’t such strong signals in the OIS forward curve six months or a year ago. This means that the market is starting to believe that the business cycle may be on its last legs. Note that even the FOMC projects a slowdown of GDP growth from 2.7% this year, to 2.4% in 2019 and 2.0% in 2020. In the longer run, the Committee projects 1.8% growth. However, financial markets may be betting on a much more pronounced slowdown. The slope of both the OIS forward curve and the Treasury yield curve sets us up for a very interesting 2019.

Figure 5: Probability of US recession at 17 month horizon is 27%

Source: Rabobank

Figure 6: Downward sloping 1m forward OIS curve on 2-6 year domain

Source: Bloomberg, Macrobond, Rabobank

Labour market indications

Signals about elevated recession risk are found not only on financial markets. In the real economy, it is especially the deviation from historical patterns that grasps our attention. Earlier, we have stated that the US labour market is closing in on historic levels of tightness (see here). Another approach here is to use so-called Okun’s Law (Okun, 1962), which is used by many economists to translate the impact of output fluctuations to unemployment rates. According to this rule of thumb, a 2 to 3 percentage point loss of real output would result in an increase of unemployment by 1 percent point. Okun’s Law has received quite some critique as it lacks a theoretical underpinning and is too static to produce proper forecasts. Therefore, we have adjusted the model by introducing three dynamic elements (see the Annex). This provides a solid model which we can use to produce unemployment forecasts based on three GDP scenarios: 1) our baseline, 2) a small recession and 3) a severe recession. The results are illustrated in Figure 7.

Strikingly, our baseline would result in unemployment rates leveling off to the natural rate and continue an ongoing spell of low unemployment rates (pink dotted line). If we plug in the US GDP figures that reflect both a mild recession (similar to the dotcom recession in 2000 and 2001) and a severe crisis (GFC in 2008 and 2009), we arrive at a completely different picture, which would be much more in line with patterns seen over half a century of labour market data.

Figure 7: Three scenarios for unemployment

Source: BLS, Macrobond, Rabobank. Note: the model up is based on quarterly data up until 2018Q1.


In this Special we have looked at risks of a US recession, the wolf that haunts our backyards once in a while and kills our flock. First and foremost, it’s impossible to predict the exact evening that you should be on guard with a shotgun, but we can assess whether or not the wolves will grow hungrier and indicate when you should look out of the window on a more frequent basis.

Our treasury yield curve model suggests that the odds of a recession happening within 17 months currently stands at 26.5%. In combination with a downward-sloping OIS curve in the 2-6 year domains, we are at least hearing a couple of stomach growls. In contrast, the Fed puts the probability of a recession occurring within 12 months at roughly at 11%. Hence, it seems to believe the wolf will continue its hibernation (which a wolf doesn’t do) with his belly full of meat. After 35 consecutive quarters of economic expansion and well on track to the break the all-time record of 39 quarters that was set in the 90s, some might even believe the wolf won’t show up anymore, but we should all know better.

Historic labour market patterns also seem to tell us something about the regularity of the wolf’s digestion pattern. From this perspective, continuing economic expansion would result in an ongoing spell of low unemployment. This could be the case if the wolf suddenly changed his diet from regular food to healthy shakes, but we seriously doubt that this is the case. Based on our analysis, the recession probability is too small to change our baseline forecasts, but substantial enough to say: wolf, wolf, wolf.

Annex: an unemployment model for the US

Okun’s Law

The use of Okun’s law is not undisputed. Some argue that the approach misses a theoretical foundation, is generally too static and the relationship breaks is unstable over time (e.g. Meyer and Tasci, 2012; Owyang and Sekhposyan, 2012). In this study, we argue that Okun’s Law is still a pretty powerful tool if dynamic elements are properly added to the specification. This is in line with research by both Gordon (2010) and Knotek (2007).

Model and data

Our model has the following form:

Where U is the unemployment rate, y is the volume of gross domestic product (GDP), yc is the volume of potential GDP, U* is the equilibrium unemployment rate, DGFC is a dummy variable for the Global Financial Crisis between 2008-2009, DICT is a dummy for the dotcom crisis in 2000-2001, D is a dummy capturing year-specific fixed effects and Δ indicates first differences. We use three dynamic elements in our model specification. First, the impact of the deviation of growth from potential growth (GDP growth gap) is disentangled in two components, where α captures the impact of high volatility on the labour market in either a downturn and β captures the impact if the economy recovers or produces according or above its potential. The threshold value to disentangle both regimes is determined by θ, which we set at -0.7. This value gives us a sufficient amount of observation in case of economic busts periods. The second dynamic element is that we use a distributed lag structure with respect to the GDP growth gap in a similar fashion as Owyang and Sekhposyan (2012). We use the actual, one quarter lagged and two quarter lagged GDP growth gap in our estimations. As a third and final dynamic element in our model, we use the insights from a study by Robert Gordon (Gordon, 2010), who transformed Okun’s Law into an error-correction model, which takes into account the long-term steady state of the labour market (U minus U*).

We use quarterly data from 1970 until 2018Q1 to estimate equation (1) using an OLS estimator. The description of the data sources is shown in Table A.1[1].

[1] We use CBO data of the natural unemployment rate, but have adjusted the data in 2016 – 2019 to match the results by Figura and Ratner (2015), who argue that the natural rate of unemployment could be at as high as 4.3% (instead of the 4.6% projected by the CBO). Figura and Ratner argue that declining union bargaining power has driven down the share of labour compensation, which gives more room to create more jobs on a structural basis.

Table A.1: Description of variables and data sources

Source: Rabobank


Table A.2 shows our estimation results. In column (1) we use the disentangled impact of the GDP growth gap, but abstract from a distributed lag structure of the GDP growth gap variables. Moreover, we have not incorporated year-specific fixed effects in this model. The estimation results show that in busts periods (α1), the impact on unemployment is far more substantial than under ‘normal’ economic conditions (β1). The magnitude of α1 is 0.36, which means that if the GDP growth gap declines by 1ppt (and GDP growth falls 0.7ppts below potential GDP, unemployment rates go up by 0.36ppts. The error-correction parameter (ECM) is statistically significant, but the coefficient is only 0.06. This means that a deviation of unemployment rates from the equilibrium unemployment rate is made up in each quarter by 0.06ppts. The slow adjustment process can be partly explained by the fact that the error correction parameters ‘struggles’ with trend-related time-specific effects. In column (2), therefore, we include year-specific fixed effects. This solves two problems that we coped with in estimations in column (1). First, including these fixed effects obviates the touch of serial correlation, indicated by the improvement of the Durbin-Watson statistic. Second, the ECM is much more robust, which results in a much faster and realistic market-clearing process than in the first estimate. In column (3) we estimate our final model by introducing the distributed lag structure. There is indeed a recursive and diminishing impact of GDP growth on unemployment over time. In addition, we have added two recessionary dummies to pick up somewhat more of the high unemployment amplitudes in recessionary times. The recessionary dummies show that the knock-on effect on unemployment during a GFC type of crisis is twice as large as with a ICT type of recession.

Table A.2: Estimation results

Notes: t-statistics are in parentheses. HAC standard errors and covariance. Significance at 10% and 5% is indicated by * and ** respectively. Source: Rabobank

Figure A.1 shows the fit of the model illustrated in column (3). The fit is accurate given the high volatility of the series. We are using this model to run three scenarios.


Our dynamic unemployment model based on Okun’s Law can also be used to make unemployment predictions for the US. We will use the model to run three scenarios:

  1. A baseline scenario
  2. Mild recession
  3. Severe economic crisis

In all three cases, we use the CBO projections of both potential GDP growth and natural unemployment rate (see footnote 1). The variable that is altered in the scenarios is GDP growth (see figure A.2). In our mild recession scenario we copy GDP growth figures seen during the dotcom recession in 2000 and 2001. For our severe crisis scenario, we use the economic decline during the Global Financial Crisis in 2008 and 2009. Besides varying GDP growth figures, we plug in recession dummy’s accordingly to pick up any additional impact on unemployment.

Figure A.1: Unemployment model is accurate

Source: BLS and Rabobank

Figure A.2: GDP scenarios

Source: BEA, Rabobank


Erken, H.P.G. and S. Koopman (2018), US: Is there rising wage growth on the horizon?, Rabobank.

Federal Reserve (2017), Changes in US Family Finances from 2013 to 2016: Evidence from the Survey of Consumer Finances. Federal Reserve Bulletin, 103, 1.

Figura, A. and D. Ratner (2015). The labor share of income and equilibrium unemployment. FEDS Notes, June, 8.

Gordon, R.J., 2010, Revisiting and rethinking the business cycle. Okun’s law and productivity innovations, American Economic Review Papers & Proceedings, 100, pp. 11-15.

Knotek, E. (2007). How useful is Okun's Law, Economic Review, vol. 92(4): 73-103.

Marey, P. and E. Gournay (2016), Is the next US recession visible in the yield curve?, Rabobank.

Marey, P. (2017), The flat yield curve and the flat Phillips curve, Rabobank.

Meyer, B. and M. Tasci, M. (2012). An unstable Okun’s Law, not the best rule of thumb. Economic Commentary, 7.

Okun, A. (1962). Potential GNP: its measurement and significance, Reprinted as Cowles Foundation Paper 190.

Owyang, M.T. and T. Sekhposyan (2012). Okun’s law over the business cycle: was the great recession all that different?. Federal Reserve Bank of St. Louis Review, 94.

Wolff, E.N. (2017). Household Wealth Trends in the United States, 1962 to 2016: Has Middle Class Wealth Recovered?. National Bureau of Economic Research, no. w24085.