This post was written by Simon van Norden of HEC-Montréal.
Brad DeLong and some others are publicly worrying about the
risks of a double-dip recession in the US as federal fiscal stimulus winds
down. Noting that
Goldman Sachs … put the odds of a technical double-dip
recession [in 2010Q2] at 25%-30%.
Brad describes this as "alarmingly large." Being a quantitative
sort of guy, I want to understand what that means.
To help, I'm enlisting the aid of the Survey of Professional Forecasters (SPF). Every
quarter since 1968, the survey has asked professional forecasters to estimate
the odds that the economy will contract in a given quarter. For example, two
years ago in 2008Q3, these professionals put the odds of a contraction in that
quarter at 34% on average. Looking back with hindsight, we now estimate that US
real GDP shrank 0.7% that quarter. So I was curious to know how things look from
the Survey's perspective.
The first thing that I realized after looking at the data was that,
technically, we should be talking about a triple-dip
recession. The BEA estimates that real GDP shrank by 0.2% in 2008Q1, grew
0.4% in 2008Q2, and then shrank for the next four quarters before starting to
grow again in 2009Q3. It's not uncommon for growth to fluctuate around zero
during periods we commonly think of as "recessions." For example, the
US economy was officially in recession from 1948Q4 to 1949Q4, but growth
was positive in 1949Q2; meanwhile growth was negative for the first two
quarters of 1947 but it was not declared a recession. Closer to home, the 2001
recession is dated from March through November of that year, but growth was
positive in 2001Q1 and Q3.
The latest data from the SPF (taken in 2010Q2) put the average odds of a
recession in Q2 at 7.4%, and in Q3 at 9.8%. That's much lower than the Goldman
Sachs odds, but its based on perceptions from about 2 months ago. (The FRB
Philadelphia will announce the results of the Q3 survey August 13th.) To put
those odds into perspective, note that 7.4% is slightly below the median of
survey results since 1968 for the current quarter. Looking one quarter ahead,
9.8% is quite optimistic; forecasters have been more pessimistic 78% of the
time.
But context might be important. In particular, I'd like to know how these
forecasters do in the aftermath of recessions. The last two times the US
economy had 2 or more quarters of negative growth in a row were in the last
three quarters of 1990 and the last half of 1981. The latter was the second
(and worse) half of a double-dip; the economy had already contracted in the
first half of 1980 and in 1981Q1. Yet after the 1990 recession, forecasters
were on average much more worried about a double-dip than they are now; in
1991Q4 they put the odds of a contraction at 33%; in 1992Q1 it was up to 43%.
Comparable figures for 1982Q3 and Q4 were 21% and 33%. Both times, growth
stayed positive.
More generally, there have been 24 quarters of negative growth since this
survey question started; if I look at forecaster's odds 4 quarters later,
they're quite gloomy on average, putting the odds of negative growth at less
than the current 7.4% on only 4 occasions (all in the 1970s.) If anything, they
seem unrealistically gloomy. 4Q after negative growth their average odds of a
contraction are 31%; contractions occurred 25% of the time. In some other recent work with John W. Galbraith, we've found that
this appears to be a pattern; in the aftermath of recessions, forecasters seem
to systematically overestimate the odds of a contraction.
Now, in writing this, I was tempted to add that, maybe, "This Time is
Different"; we've had a nasty financial crisis and growth in Europe looks
esp. weak. Then I thought back to 1992…..and remembered the financial crises
in the EMS that summer and next….and thought back to 1982….and remembered
the collapse of world commodity prices and the Latin American debt defaults. So
I think its useful to remember that those were very, very turbulent times as
well.
So what should a macro-wonk conclude from all this?
·
Compared to what other forecasters have been
saying, Goldman-Sachs' 25% odds look very high.
·
Most forecasters think downside risks are low on
average and low for an early recovery phase.
·
These same forecasters tend, if anything, to
overestimate the downside risks after a recession.
The caveat I'm keeping in my mind is
·
The SPF numbers were released two months ago and
the data have not been strong since; I'm interested to know what the new
numbers will look like in four weeks.
Worthwhile Canadian Initiative 
“These same forecasters tend, if anything, to overestimate the downside risks after a recession. ”
Nice application of risk-neutral vs actual probabilities perhaps?
Recessions are a time of/caused by increases in risk premiums.
So, forecasters’ expectations are a lagging indicator? 😉
No, they overweight bad outcomes.
The “risk-neutral” probability of a state is the physical probability of the state multiplied by the marginal utility of consumption in that state. Bad states (low consumption states) get a higher weight than good states because the marginal utility of consumption is higher in those states (you value an extra unit of consumption more if you already have less of it).
Adam P.
Why would a forecaster report a risk-neutral density?
– They claim to report an objective density.
– Whose risk-neutral density is it? Remember, I’m looking at the average probability across forecasters. The assumption that they’ve all using the same weighting is …. um … a very strong one.
– Why is the weighting that they’re using seemingly different during recoveries as opposed to, say, at recession troughs, or times of highest uncertainty?
Believe what makes you happy, but I don’t find the risk-premium story compelling for professional forecasts.
Simon, nobody means to report a risk-neutral density. It may not be that anyone consciously prices according to a risk-neutral density either.
This:
“-Whose risk-neutral density is it? Remember, I’m looking at the average probability across forecasters. The assumption that they’ve all using the same weighting is …. um … a very strong one.”
is a strange and irrelevant comment. The physical forecast densities can all be different and the risk adjustments all different, where do you get the idea that I was assuming they all used the same weighting? Do you teach your classes that prices can only exist if everyone uses the same subjective risk-adjusted probability weighting? Scary.
“- Why is the weighting that they’re using seemingly different during recoveries as opposed to, say, at recession troughs, or times of highest uncertainty?”
I don’t know but the degree of uncertainty is not that relevant here, it’s the magnitude and direction of the risk bias. Yes, if it’s risk aversion in the usual sense then you’d expect a pessimistic bias in the trough of the recession too. However, it does depend on what constitutes the “bad” states at different points in the cycle.
Suppose it is a particularly bad mistake for a professional forecaster to over-estimate a recovery (somehow, in terms of reputation and career advancement). Then the rational response would be to bias down your forecasts in those times, that would in fact be an example of risk-adjusting your probabilities.
Anyway, believe what makes you happy but if you don’t have a better response then that I hope you don’t have to teach anything to do with risk-neutral pricing.
Click to access aer_98_2.pdf
From the paper:
“Indeed, the fact
that the coefficient on the FOMC forecast is
negative (albeit not significantly so) suggests
that someone trying to forecast inflation shouldmove away from the FOMC forecast, not toward
it. Taking into account the likely heteroskedasticity
and serial correlation of the residuals only
strengthens the results. The WLS estimates with
robust standard errors show that the weight on
the staff forecast in predicting inflation is 1.40
1t 5 5.542 and that on the FOMC forecast is
more negative and close to significant.”
The people who actually set policy tend to overestimate inflation relative to the staff forecasts. To the extent that the members of the committe see inflation above what they intend as a particularly bad outcome (and we are learning right now that they do) then this is a clear example of risk adjusting their probabilities. (The data on this paper would precede the current crisis, sample ends in 2001).
When Krugman and De Long call for a double-dip recession which definition of recession are they using? The one set by the NBER committee, or the typical 2 consecutive quarters of negative growth?
Or is one quarter of negative growth sufficient for a “dip”?
Westslope: I read the above quote about the GS forecast to mean a contraction in real GDP in 2010Q2. Note that growth was positive in Q1, so this doesn’t meet many definitions of recession.
Adam P: “Nobody means to report a risk-neutral density. It may not be that anyone consciously prices according to a risk-neutral density either.” I’m an empirical guy, Adam. I try to listen to data. I don’t start my day asking “how can I reinterpret the world to fit the way I think it should work.” And I try not to argue about things that are observationally equivalent. And I teach option pricing.