Transcript Discussion Slides

Credit Spreads and the Severity
of Financial Crises
Krishnamurthy and Muir
Discussion by Robin Greenwood
Main Idea: Crisis Research 2.0
• Many of the empirical papers that either
forecast crises or uses crises to forecast X are a
bit ad hoc
– Selecting crisis dates in particular involves judgment
• Supplement crisis data with (impressive data
on) credit spreads from historical sources
• Can this data help us say anything about crises?
– Yes! Cross-sectional variation helps
– Helps sort crises ex ante as to which ones will be
particularly bad
– Helps look at role of expectations
Position of the Financial Sector
Main Empirical Innovations
• GDP(t+1) = a + b*CrisisDummy(t)*Spread(t)
• Cross-sectional variation in the spread across
crises is useful for forecasting impact of the crisis
• Also shows that spread crises preceded by credit
growth turn out poorly for subsequent GDP, but
unclear whether the credit spread adds anything
here
– Does not run horse race compared to traditional crisis
variables
Table 3
• Credit spreads are mainly useful in forecasting
output during financial events
• During garden variety recessions, the credit
spread is not particularly useful
• Poses a bit of a puzzle in terms of the forward
looking nature of the credit spread
Which Spread Crises Turn Out Badly?
• Table 11
• Regress GDP growth on dummy for SpreadCrisis
and SpreadCrisisxLaggedCreditGr
• Bottom panel just uses credit growth with nearly
identical results
• Is the paper saying that we can ignore whether a
financial crisis actually happened?
• I would have liked a bit more nuance here,
perhaps showing quantile regressions
– Most of the time, credit growth is good for GDP
growth, except when it leads to financial crisis
Comment 1: Timing of the Credit Spread
Crisis Forecasting & the Credit Spread
Forecasting is Hard
• In the future, when financial crises occur, we will
be watching them in real time
• Which credit spread will be relevant for
forecasting GDP? When should we measure it?
– At the onset of the crisis?
– 6 months in?
– Data from the last crisis show that this is not at all
obvious
– Paper is not very clear about exact timing of this
measurement
Comment 2: Value Added of Incorporating
Credit Spreads
Credit Growth and Credit Spreads
• Paper shows very clearly that credit spreads help forecast impact of
crisis on economy
• Less clear that there is much new in the results the paper shows using
credit spreads
• We know from many other papers that credit growth is associated
with
– Financial crises (Schularik and Taylor)
– Low subsequent returns on credit (Greenwood and Hanson) and equities
(Baron and Xiong)
• Paper does not show very persuasively that using credit spreads
increases the information content of credit growth for forecasting
outcomes
• Would be helpful to show some horse races (like in Table 7, but using
GDP growth as the dependent variable) with crisis measures vs. the
authors’ augmented crisis measures
Forecasting Power
• Crisis forecasting literature often focuses on Rsquared concept, on which the paper is largely
silent
– How much of an edge are credit spreads giving us
in forecasting?
Comment 3: Forecasting the Crisis vs. The
Impact of the Crisis Conditional of it Happening
Paper’s perspective on expectations
• Crisis itself is (not very) forecastable, but high credit
growth before the event leads to fragility, so that if a
crisis unfolds, credit spreads quickly change
“It occurs at first very slowly, then all at once.”
- Ernest Hemingway describing the process of going broke
• I would have liked help in separately unpacking the
– Forecastability of the crisis itself
• Is the “surprise” predictable?
– Fragility of the crisis
• Results in Lopez-Salido, Stein, and Zakrajsek (2015) suggest that
it is the forecastability of credit spreads, rather than fragility,
that is the driving factor forecasting GDP in the US
• Could easily be extended using the authors’ database
My Preferred Crisis Narrative
• Periods of credit growth are driven by investor
optimism
– Extrapolation of GDP growth; extrapolation of past
default rates
– One result is misallocation of credit
• Higher credit growth forecasts low returns on
average
• In some particularly severe incidents, losses
affect financial institutions, resulting in a crisis
• Can the paper reject this narrative?
Are Credit Spreads too low?
Are credit spreads too low?
• A more natural way to capture this is using
return forecasting regressions
– Greenwood and Hanson (2013)
– Lopez-Salido, Stein, and Zakrajsek (2015)
• Suppose we start with baseline view that
returns to credit are constant over time
• Then “forecastable part” of returns is then the
amount that credit spreads are too low
Introduction
Empirical Strategy
Issuer Quality
Forecasting Results
Interpretation
Summary
Quantity vs. Quality

Table 4: Panel A
ISSEDF
rxtHY
2  a  b1  X1,t  b2  X 2,t  ut 2
-15.254
[-5.29]
DAgg/DAgg (Agg. debt growth)
-12.978
[-3.78]
-5.212
[-3.97]
-2.433
[-1.49]
D1/D1 (Low EDF debt growth)
-3.474
[-2.04]
D5/D5 (High EDF debt growth)
-1.565
[-0.86]
-7.091
[-3.76]
-6.631
[-3.09]
D5/D5 - D1/D1 (High-Low)
R2
Quality beats
Quantity in a horserace
→Similar results for HYS
0.26
0.13
0.29
0.06
Credit growth of
low quality firms is most
useful for forecasting returns
0.24
-4.917
[-3.16]
0.26
-5.420
[-2.39]
-6.538
[-3.09]
0.14
0.26
Differential debt
growth of low vs. high quality
firms is a strong predictor
Open Questions
• The calm before the storm
“It occurs at first very slowly, then all at once.”
- Ernest Hemingway describing the process of going broke
• Credit growth vs. Credit Levels
• The role of expectations in driving the crisis in
the first place
“The US housing marketplace developed an axiom over time.
Since housing prices had never fallen, housing prices could
therefore never fall. This is exactly how repetitive events,
whether it’s housing moving higher year after year, or JGBs going
higher in price every single year, become cognitive biases. Then,
at some point in time, they actually become axiomatic. These
axioms are rooted in belief systems that are developed through
inductive reasoning based upon repetition of perceived fact
patterns. When these patterns move to excess and can no
longer be sustained, they reverse themselves in a violent
manner.”
– Kyle Bass (investor short subprime)