That’s the title of a new FRBSF Economic Letter. From Future Recession Risks, by Travis Berge and Oscar Jorda:
An unstable economic environment has rekindled talk of a double-dip recession. The Conference Board’s Leading Economic Index provides data for predicting the probability of a recession but is limited by the weight assigned to its indicators and the varying efficacy of those indicators over different time horizons. Statistical experiments with LEI data can mitigate these limitations and suggest that a recessionary relapse is a significant possibility sometime in the next two years.
The key graph is here:
Figure 3 from Berge and Jorda (2010).
Taking out the yield curve leads to a much higher implied probability of recession, based upon historical correlations.
From the Conclusion:
Any forecast 24 months into the future is very uncertain. At two years out, the odds of recession vary
from almost three times more likely than expansion, to expansion being almost five times more likely
than recession, depending on which LEI components are used. Nevertheless, LEI forecast trends
indicate that the macroeconomic outlook is likely to deteriorate progressively starting sometime next
summer, even if the data suggest that a renewed recession is unlikely over the next several months. Of
course, economic policy can strongly influence the outcome. The policies that are adopted today could
play a decisive role in shaping the pace of growth.
This confirms my views that (1) the yield spread will lead to misleading inferences in the current environment, as suggestesd in Kucko and Chinn (2010) [1], and (2) in light of the weakness in economic activity [0] our elected leaders need to recognize the hazards confronting the economy, and implement additional stimulus [2].
I’ll have a post up soon on why we don’t need to worry about inflation, in the short to medium term.
Poor choice of words, I’m afraid. The authors run a whole bunch of series, and the one that shows highest odds of a recession is the one without the yield curve. That does not “confirm” that the yield curve leads to misleading inferences, unless you have knowledge that recession odds are actually as high as that particular series indicates. That series agrees with your expectation, perhaps?
The problem here is that, with odds of recession near (at?) zero in the near term, there is only one direction for odds to go, as time passes. So we see odds of recession rising over time. Isn’t that the case in any non-recessionary time? Time passing allows more time for things to go wrong.
This study is getting lots of blog coverage, but is it really deserved? Isn’t one way to look at the results that, if one runs combinations of data, some combinations will show higher odds of some result than others. If odds start out near zero, they will certainly show an increase as time passes. Do we really know any more about the economy than we did? Isn’t there a good chance, the way this was constructed, that the outcome would have been similar based on sorting randomly chosen variables?
I’d add 3rd point – government needs to recognize the negative impact that an uncertain policy environment has on hiring.
A variety of proposed and existing payroll taxes make it difficult for employers to ascertain the degree to which marginal value product of hiring additional workers offsets the cost of those new hires. Constitutional questions of some policies(I’m thinking health care and potential EPA mandates) implies that the level of uncertainty will not be resolved for more than a year while we wait for the Supreme Court to rule.
In anticipation of the blame Bush arguement, or blame the Republicans arguement, I want to point out a relevant aspect of the debate. That is, any time a party introduces policies that push the envelope of constitutionality, it should raise a red flag. I don’t care which party it is, but the other needs to be responsible enough to call them on it and when necessary, let the court decide. That’s about all the Republicans could do in the current debate. As we saw, it did little good for Republicans to propose alternate solutions because Pelosi, Reid and the democratic majority in both houses/committees prevented any real debate. Obama and the White House pushed bribes for votes and the result is legal challenges to the policies of the fundamental changers.
Let’s hope we get enough conservatives and moderate democrats elected in the fall so that congress can steer policy like they did in the Clinton years.
kharris: If the probability is near zero, one could still stay near zero — if the underlying indicators continue to rise, for instance.
I agree that it would be better to say “is consistent with”. I think the yield curve leads to misleading inferences right now, because in Kucko and Chinn, we find that yield curve does disasterously poorly in predicting subsequent growth since the trough (although it did do well predicting the recession. But that prediction was driven by the sample predating the Fed funds rate hitting zero).
I’ll venture a guess that a higher percentage of economists are employed by governments, than say engineers or salesmen. So in a government stimulated “recovery”, it may look to some economists that there is a recovery and a “future recession” is something to analyze. For those of us who are employed in private industry, there has been no recovery. Hours are down, wages are flat with anecdotes of decline, private employment was recently announced as essentially flat, unemployment is 10%, underemployment is 20%, unemployed beyond 26 weeks is hitting records, house prices remain down with hints of further decline, new housing starts are at record lows, several banks and a car company remain owned by the government and taxpayers are paying to keep them in operation, etc.
In private industry, we do not call this a recovery. There is no need to talk about a second “future recession”, we are still in the first one. I guess it all depends on which end of the telescope one looks through.
If there’s a recession coming, I don’t think we’ll have to wait two years to see it.
If the statistical results are based on the post-war U.S. experience, they are not worth much, as we have not experienced a balance sheet recession in that time. The correlations with the yield curve (made uselss by the liquidity trap) are not the only ones that need to be thrown out
Menzie: When you looked at the German yield spread and output data, how did you handle industrial output during the Reunification period? What little I’ve read of German macro data (and it is very little), it seems that they almost always assume a regime change in the data during the Reunification period. Or was it just not an important consideration in this case?
I’ve been doing work with John W. Galbraith of McGill University on, among other things, the Survey of Professional Forecasters Recession Probability forecasts and the Bank of England GDP Fan Charts. In both cases, we look at the estimated probability of negative GDP growth from 0Q to 8Q into the future. In both cases, we find broadly similar results; for forecasts more than about 2Q ahead, variations in the probability of recession are basically meaningless. Historically, such changes do not correlate with variations in the frequency with which we observe recessions. This seems to line up well with evidence others have found on the difficulty of predicting negative GDP growth.
I’m therefore wondering about that 2yr recession probability prediction. How much evidence is there that this LEI works at those kinds of horizons?
Menzie,
If the independent variables were randomly selected, with no reason to think they were any good at predicting the dependent variable, we’d expect to see results similar to the ones we got. Starting near zero in a dependent variable bounded by 0% and 100%, we’d expect to see the dependent variable rise. So yes, if these are truly leading indicators, they are showing a rise in recession odds over time, and if they all continued to rise, they might not show rising recession risk.
However, we are told at the outset that, organized as the Conference Board has organized them, these “leading” indicators don’t lead. If they are organized in a variety of other ways, they can be made to show rising recession odds over time, just as a bunch of randomly selected independent variables would do. This doesn’t strike me as a very stringent test.