+10 votes
asked ago in Current Economic Issues by (1.2k points)
The US expansion has lasted for 106 months, leading many to worry that it has to end soon.  I know of two lines of work.  One dating back to the work of Rudebusch and Diebold (for a relatively recent update in 2016, link below) and finding  little to no effect of the length of the expansion on the probability of a recession.   The other, rehabilitating Minsky, argues that, as the expansion lasts, risk increases, leading inevitably, and with increasing probability, to a crash and a recession.  I however do not know of conclusive empirical evidence that this is indeed the case.   I thought this would be a good question to ask in this forum.  

commented ago by (3.3k points)
Australia says

"Hold my beer, mate"
commented ago by (150 points)
Durland and McCurdy (1994 JBES) developed an elegant framework for answering this question; the duration-dependent Markov-Switching Model. They found no empirical evidence of a link between expansion length and the probability of transition from expansion to recession.
I'm not aware of updates to their results.

6 Answers

0 votes
answered ago by (2.7k points)
edited ago by
The faster the full restoration of the economy the faster it collapses again (When is not a exporter country). With more inflation related to stable costs and smaller injection into the economy there is a deeper crisis.
0 votes
answered ago by (1.8k points)
I also am not aware of any empirical evidence suggesting this is the case. Asymmetry in the speed of collapse when the economy collapses, yes. Increasing probability of collapse with the length of the expansion, no. Neither the pre-1929 nor the pre-2008 expansion was notably long.

Why suppose that risks accrete slowly over time? Why not pursue the line of inquiry that risks are generated primarily by positive-feedback trading episodes, which do not move slowly?

Brad DeLong
0 votes
answered ago by (3.5k points)
https://ntguardian.wordpress.com/2018/04/19/a-recession-before-2020-is-likely-on-the-distribution-of-time-between-recessions/ gives an interesting blog post modeling recessions with a Weibull. The suggestion is that recession is relatively likely to occur soon.
commented ago by (1.8k points)
Interesting... But remind me why we should think the 1950s, 1960s, and 1970s are relevant to estimating expansion durations in the current expectational-policy régime? If they are not, we have not 13 but 3 datapoint—the 1980s expansion, the 1990s expansion, the 2000s expansion. The 1980s expansion was brought to an end with a (small-to-moderate) recession because inflation had crept up to 5% and the S&L crisis hit; the 1990s expansion was brought to an end with a (small) recession because of the collapse of the dot-com bubble. The 2000s expansion was brought to an end by the extraordinary concentration in big banks' portfolios of AAA-rated assets that were not truly AAA at all.

We have observations of 9 years, 8 years, 7 years. Conditional on an expansion having already lasted 8 years, we have an expected duration of 1 more year and... no estimate of the variance of the expected duration at all...

Brad DeLong
commented ago by (3.5k points)
If the process for the business cycle has largely changed, then we have essentially no data and--as you say--there isn't going to be much statistical evidence on the question.

On the non-statistical side...the Great Recession happened because we allowed really stupid behavior. Arguably, the dot-com crash illustrated really stupid behavior. I think the S&L crisis definitely illustrated really stupid behavior. (All of these somewhat  centered around financial markets.)

Would you have an opinion on whether there is any really stupid behavior going on in the economy at the moment...
commented ago by (700 points)
FWIW here's a histogram of the length of recoveries in months for 33 OECD countries + 6 other large economies (all the countries for which the OeCD provides business cycle dating).

(#1) https://imgur.com/a/cy7NSdW

It looks pretty indistinguishable from an exponential distribution in which the probability of the recovery ending is constant each month
  • (#1)
commented ago by (1.8k points)
interesting. Thanks...

Brad DeLong
commented ago by (3.5k points)
Here's a piece of evidence in support of Brad's suggestion that the most recent three recessions are really different. Apparently a drop in conceptions forecasts a coming recession for the three most recent recessions but not earlier. Source is a nice paper by Kasey Buckles and coauthors, https://www.nber.org/papers/w24355.pdf.
+1 vote
answered ago by (240 points)
The evidence, as far as I know, suggests that there is no systematic link between the length of an expansion and the probability of a recession.

But this, in and of itself, does not rule out that Minsky-type hypotheses are valid. Indeed, there is plenty of evidence that prolonged credit growth, especially if alongside property price increases, provides useful leading signals of banking crises in real time (for one of the many examples, see here   https://www.bis.org/publ/qtrpdf/r_qt0903e.htm ). Since banking crises are costly for output, it stands to reason that such indicators should also work as useful predictors of recessions. I fact, recent work we have done bears this out (it is forthcoming in the next BIS Quarterly Review, out in December).

There is also evidence that the financial expansions and contractions (financial cycles) most damaging for output tend to be longer than business cycles, as traditionally measured (eg, here: https://www.bis.org/publ/work380.pdf ). This is naturally consistent with more frequent downturns (or ends to a business cycle expansion). For instance, an outsize financial cycle straddled the mild 2001 recession in the US, before ending in a much more damaging bust that coincided with the Great Financial Crisis and recession. From this perspective, one could consider the 2001 downturn an “unfinished recession” (see same source above). This evidence is consistent with the idea that risks build up gradually during expansions.
To summarise: conditioning on the age of the business cycle expansion does not help predict recessions; conditioning on outsize financial expansions does. Risks builds up in a financial boom, and materialise in the subsequent bust.
commented ago by (1.8k points)
Very true and smart: "conditioning on the age of the business cycle expansion does not help predict recessions; conditioning on outsize financial expansions does".

Cf. also the excellent: **Òscar Jordà, Moritz HP. Schularick, and Alan M. Taylor** (2014): _[Betting the House](https://www.nber.org/papers/w20771)_: ".Is there a link between loose monetary conditions, credit growth, house price booms, and financial instability? This paper analyzes the role of interest rates and credit in driving house price booms and busts with data spanning 140 years of modern economic history in the advanced economies. We exploit the implications of the macroeconomic policy trilemma to identify exogenous variation in monetary conditions: countries with fixed exchange regimes often see fluctuations in short-term interest rates unrelated to home economic conditions. We use novel instrumental variable local projection methods to demonstrate that loose monetary conditions lead to booms in real estate lending and house prices bubbles; these, in turn, materially heighten the risk of financial crises. Both effects have become stronger in the postwar era...
commented ago by (110 points)
The fact that so many people have continued to doubt this recovery at every step of the way makes Minsky-like dynamics less likely to emerge. That's what has me worried about the "Trump bump" - the tax cuts and the promise of deregulation convinced a lot of businesses that the good times are here to stay. Strong confidence is great in the short run, but we may pay for it once the stimulus fades and it becomes clear these policies haven't launched the US on a permanently higher growth path. That said, there aren't any obvious accelerants for the downturn this time around, except for arguably little monetary or fiscal policy slack, so I'd expect the "Moderate Moderation" to continue with only a slight interruption.
+1 vote
answered ago by (1.2k points)
Trying to distill what I learned from the exchange.   

The graph given by Anna  Stansbury is rather convincing:  Conditional only on duration, the probability that an expansion comes to an end seems roughly constant
 (but not quite: one sees more bars above the exponential for short durations, and more bars below the line for long durations.   Would we worth a formal statistical test.  _

The point by Claudio Borio is also rather convincing,.  ``Credit booms'', i.e. sustained credit growth seems more likely to lead to financial crises, which presumably often lead to the end of an expansion.   

This raises the obvious issue:   The relation of sustained expansions to credit booms.  Is it the case that the longer the expansion has been going on, the higher the probability of a credit boom?   (Or put another way, what is the empirical relation between the ``business cycle'' and the ``financial cycle''? )
+3 votes
answered ago by (200 points)
Maybe there is a different way of thinking about an aging recession which is not just by measuring time. We know that as expansions get mature unemployment goes down. As unemployment reaches a low level, is the likelihood of a recession increasing? I think the answer to that question for the US is a clear yes.

Using quantile regressions one can show that the tail risk (probability of a large increase in unemployment) increases fast as unemployment decreases. Another way to illustrate the same conclusion is the fact that unemployment rates in the US have a clear V shape around all recessions. Interestingly, this is not true for other countries where unemployment can be flat for years after it reaches a low level. And quantile regressions for those countries do not show a clear increase in tail risk as unemployment goes down. One gets a sense that in the US "full employment" is not a sustainable state (just to be clear, there are other countries that look like the US but not all of them do)

This can be, of course, related to the argument of Claudio Borio, that credit or other financial imbalances grow over time. But it would seem that in the US case, there is always an unavoidable imbalance that appears as unemployment reaches low levels, which is not true in other countries. So mature expansions (as measured by low unemployment rates) are more likely to die.

I put some of these results in a short note in case anyone is interested in the details:

As a caveat: there are not many recessions so it is always hard to establish a rule that will apply to all future recessions. But this is what the data we have shows.
commented ago by (1.2k points)
Antonio is clearly right.  When unemployment is very low, growth cannot continue as it did during the recovery, and the risk that it turns negative must be higher.  (Very nice note).  The issue is the role of the central bank in the process.   Is it typically the case that, as unemployment becomes very low, inflation increases, and the central bank increases interest rates, leading to a hoped for slowdown, but often to a recession as well?   Antonio: more work?  :)
commented ago by (150 points)
I interpreted the original question posed in this thread to be about economies in general rather than just the US economy.

One might wonder, if there is a shortage of turning points in the US economy, whether pooling together results from other economies would provide a more informative sample.  In particular, given the much high persistence in unemployment rates in other developed economies, I wonder whether the informativeness that Prof. Fatas documents in US unemployment rates is typical of what would be found in a broader sample.