Credit Spreads and the Severity of Financial Crises Tyler Muir, Yale University

Credit Spreads and the Severity
of Financial Crises
Arvind Krishnamurthy, Stanford University and NBER
Tyler Muir, Yale University
October 2, 2015
Three results
1. Financial crises are followed by deep and protracted recessions
• Estimates
• Comparison to non-financial recessions
2. The severity of the GDP contraction following a crisis can be
forecast based on the rise in spreads in the first year of the crisis
and the extent of credit growth pre-crisis
3. Spreads pre-crisis are abnormally low
• “Froth” precedes crises
2
Data: Credit spreads, crisis dates, GDP
• 1869-1929 across 14 countries from old newspapers
• 1930-present from various central banks and other data sets
(Datastream, Global Financial Database) for more recent credit
spreads
• High grade minus low grade corporate spread
• Corporate bond index to government bond
• We normalize each country’s spread as:
𝑠𝑝𝑟𝑒𝑎𝑑𝑖,𝑡
𝑠𝑖,𝑡 =
𝑠𝑝𝑟𝑒𝑎𝑑𝑖
• Total of 900 country-year observations
3
Credit spreads: 1869-1929
• Individual bond prices on banks, sovereigns, railroad, etc.
• Over 4000 unique bonds, 200,000 bond / years
• We convert to yield to maturity
• Spread = high 10th percentile avg yield minus low 10th percentile avg yield
4
Data: Credit spreads, crisis dates, GDP
• We cross this data with crisis dates from Reinhart-Rogoff (RR) and
Schularick-Taylor (ST) and Bordo-Eichengreen-Klingebiel-Martinez
(BE)
• Total of 900 country-year observations:
• 44 ST crises
• 48 RR crises
• 27 BE crises
• GDP data from Barro-Ursua
5
Empirical approach
• Define a set of dates identified with a major financial crisis
• Examine the behavior of output (and spreads) around these dates
• Choice of dates is important!
• What defines an event as a “financial crisis”?
6
Theory: What is a financial crisis?
• Shock 𝑧𝑡 : recessionary shock, lower expected cash-flows on assets held by
intermediaries
• Fragility 𝐹𝑡 : high leverage/low equity capital, short-term debt, correlated
intermediary positions, interconnected exposures
• “Trigger” + “Amplification”
• Asset price feedback
• Credit crunch
• Bank runs/failures/disintermediation
• Credit spreads rise:
• Expected default + risk/illiquidity premium
• Kiyotaki-Moore, He-Krishnamurthy, Brunnermeier-Sannikov
7
Result 1: Aftermath of a crisis
Crisis severity = 𝑧𝑡 𝐹𝑡
• What should one have expected,
standing in 2008?
• Reinhart and Rogoff (2009)
• Across a set of defined events:
-9.3% Peak-to-trough
• Slow recovery
8
Credit spreads are a continuous measure
• Spreads rise more in more severe crises (𝒛𝒕 𝑭𝒕 high):
𝑠𝑖,𝑡 = 𝛾𝑖 +
𝑦𝑖,𝑡+𝑘
𝛾1 × 𝐸𝑡 ln
+
𝑦𝑖,𝑡
Default, Risk,
𝛾2 𝑧𝑡 𝐹𝑡
Illiquidity
• Underlying relation: 1𝑐𝑟𝑖𝑠𝑖𝑠,𝑡 if 𝑧𝑡 𝐹𝑡 >Threshold
𝑦𝑖,𝑡+𝑘
ln
= 𝑎𝑖 + 𝑎𝑡 + 𝜷 × 𝒛𝒕 𝑭𝒕 × 1𝑐𝑟𝑖𝑠𝑖𝑠,𝑡 + 𝑐 ′ 𝑥𝑖,𝑡 + 𝜖𝑖,𝑡+𝑘
𝑦𝑖,𝑡
9
Specification
• Panel data regressions (country 𝑖, horizon 𝑘):
• Interact spreads with crisis dummies
𝑦𝑖,𝑡+𝑘
𝑐
ln
= 𝑎𝑖 + 𝑎𝑡 + 1𝑐𝑟𝑖𝑠𝑖𝑠 × [𝑏 𝑐 × 𝑠𝑖,𝑡 + 𝑏−1
× 𝑠𝑖,𝑡−1 ] +
𝑦𝑖,𝑡
𝑛𝑐
1𝑛𝑜−𝑐𝑟𝑖𝑠𝑖𝑠 × [𝑏 𝑛𝑐 × 𝑠𝑖,𝑡 + 𝑏−1
× 𝑠𝑖,𝑡−1 ] + 𝑐′𝑥𝑡 + 𝜖𝑖,𝑡+𝑘
• Controls: lagged GDP growth, 3 year credit growth from Schularick-Taylor
• Standard errors cluster by country
• Why still use 0-1 crisis dummies?
10
ST crisis recessions and no-crisis recessions
• Schularick and Taylor (2012), Jorda, Schularick and Taylor (2013)
…we define financial crises as events during which a country's banking sector
experiences bank runs, sharp increases in default rates accompanied by large
losses of capital that result in public intervention, bankruptcy, or forced merger
of financial institutions….
• Crisis date based on start of recession accompanying financial crisis
(44 events in our sample)
• Non-financial recession dates (100 events)
11
RR and BE: dates based on failure event
• Reinhart and Rogoff (2009): (48 crises)
We mark a banking crisis by two types of events: (1) bank runs that lead to the
closure, merging, or takeover by the public sector of one or more financial
institutions; and (2) if there are no runs, the closure, merging, takeover, or
large-scale government assistance of an important financial institution (or
group of institutions), that marks the start of a string of similar outcomes for
other financial institutions.
• Bordo, et. al., (2001): (27 crises)
We define financial crises as episodes of financial-market volatility marked by
significant problems of illiquidity and insolvency among financial-market
participants and/or by official intervention to contain those consequences. For
an episode to qualify as a banking crisis, we must observe financial distress
resulting in the erosion of most or all of aggregate banking system capital.
12
Standard
errors in
parentheses
13
Scatter plots: results not driven by outliers
14
Comparing GDP outcomes across events
• We plot a Jorda projection impulse response:
𝐸 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ 𝑠𝑝𝑟𝑒𝑎𝑑 + 1, 𝑒𝑣𝑒𝑛𝑡 ]-𝐸 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ 𝑠𝑝𝑟𝑒𝑎𝑑, 𝑒𝑣𝑒𝑛𝑡 ]
For crisis events, and recession events
• This attempts to mimic: underlying structural shock, 𝑧𝑡 , the same in two economies.
• Economy A: levered financial sector, financial crisis, financial recession
• Economy B: no financial crises, and we have a non-financial recession
• However, it is likely that 𝑧𝑡𝑟𝑒𝑐𝑒𝑠𝑠𝑖𝑜𝑛 for same spread change is larger than 𝑧𝑡𝑐𝑟𝑖𝑠𝑖𝑠 , which means crisis
impulse is an underestimate
15
Impulse response to +1 shock (Jorda projection)
-9% at 4yrs
-3% at 4yrs
𝐸 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ 𝑠𝑝𝑟𝑒𝑎𝑑 + 1, 𝑐𝑟𝑖𝑠𝑖𝑠/𝑟𝑒𝑐 ]-𝐸 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ 𝑠𝑝𝑟𝑒𝑎𝑑, 𝑐𝑟𝑖𝑠𝑖𝑠/𝑟𝑒𝑐 ]
16
2008 Actual and Predicted (using ST dates)
17
ST versus RR
-9% at 4 yrs
-2.7% at 4 yrs
18
Summary and some doubts
• Recessions with financial crises are deeper and more protracted than
recessions without financial crises
• -9% versus -3%, 4 years out
• But results depends on dating used
• Why date before bank failures (ST versus RR/BE)?
• Is dating subject to peek-ahead bias?
19
Result 2: Losses X fragility
Theory:
crisis severity = 𝑧𝑡 × 𝐹𝑡
• 𝑧𝑡 in most models is losses on bank assets
• 𝐹𝑡 is leverage
20
Losses ∝ change in spreads
Standard errors in
parentheses
21
5 year GDP growth
Losses ∝ change in spreads
Standard errors in
parentheses
22
Define events based on 𝑧𝑡 and 𝐹𝑡
• Shock 𝑧𝑡
SpreadCrisis = 1 𝑖𝑓
𝑠𝑖,𝑡 − 𝑠𝑖,𝑡−1 𝑖𝑛 90𝑡ℎ 𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑖𝑙𝑒
𝐶ℎ𝑎𝑛𝑔𝑒
𝐷
𝑖𝑛
𝑃
> 𝑚𝑒𝑑𝑖𝑎𝑛
• Fragility 𝐹𝑡 : 3-year growth in Credit/GDP from Schularick and Taylor
HighCredit = 1 if 3-year growth in Credit/GDP>median
23
Crises outcomes based on losses X fragility
HighCredit = 1 if 3-year growth in Credit/GDP>median
24
Impulse responses based on alternative dates
• High Credit is free of peek ahead bias
25
Result 3: Pre-crisis behavior
Lagged 3-year credit/GDP growth (ST crises)
26
Spreads pre-crisis and credit growth
• Pre-crisis, growth in 𝐹𝑡 is observable
𝑠𝑖,𝑡−1 = 𝛾𝑖 +
𝛾1 × 𝑃𝑟𝑜𝑏 𝑧𝑡 > 𝑧 𝐸𝑡−1
𝑦𝑖,𝑡+𝑘−1
ln
|𝑐𝑟𝑖𝑠𝑖𝑠
𝑦𝑖,𝑡−1
As 𝐹𝑡 rises
27
Spread pre-crises compared to other periods
28
Spreads over the cycle (ST Crises)
0 means
𝑠𝑝𝑟𝑒𝑎𝑑𝑖,𝑡 = 𝑠𝑝𝑟𝑒𝑎𝑑𝑖
29
Investor beliefs
• Pre-crisis, growth in 𝐹𝑡 is observable
𝑠𝑖,𝑡−1 = 𝛾𝑖 +
𝛾1 × 𝑃𝑟𝑜𝑏 𝑧𝑡 > 𝑧 𝐸𝑡−1
𝑦𝑖,𝑡+𝑘−1
ln
|𝑐𝑟𝑖𝑠𝑖𝑠
𝑦𝑖,𝑡−1
Pre-crisis
Pre-crisis as
𝐹𝑡 rises
• A crisis is a surprise; large shift in investor expectations
• Caballero-Krishnamurthy (2008): Knightian uncertainty
• Gennaioli-Shleifer-Vishny (2012): Neglected risks
• Not a slow build up model, where crisis prob rises over time
• Gorton-Ordonez (2014), Boissay-Collard-Smets (2014)
30
Conclusion
• Aftermath of financial crises is deep and protracted recession
• Effect of crisis lasts many years
• We use variation in severity indexed by spreads
• Results consistent with Reinhart-Rogoff, Schularick-Taylor; we give more
precise answers
• Spikes in spreads + real fragility = Losses + Amplification
• Lead to poor GDP outcomes (Kiyotaki-Moore, He-Krishnamurthy)
• Crises are preceded by unusually low spreads
• Spreads pre-crisis do not price an increase in fragility
• “Surprise” is a key dimension of crises (Caballero-Krishnamurthy, GennaioliShleifer-Vishny)
31
Extra pictures
Spreads recover quickly, GDP drop persists
33
Spreads recover quickly, GDP drop persists
Initial spread matters
many years out
34
2008 Actual and
Predicted
35