How Important is the Global Financial Cycle?

slide1. How Important is the Global Financial Cycle?Evidence from Capital Flows Eugenio Cerutti, Stijn Claessens
and Andrew K. Rose<br>
slide2. Q: How Important is Global Financial Cycle for Capital Flows? Rey and others: very
“Risky asset prices around the globe, from stocks to corporate bonds, have a strong common component. So do capital flows … The global financial cycle can be related to monetary conditions in the centre country and to changes in risk aversion and uncertainty … capital flows … are largely driven by a global factor … “
- Rey (2013, pp 1-2) Rose: Quantifying Global Financial Cycle Effect on Capital Flows 2<br>
slide3. Key Finding: Effects of Global Financial Cycle on capital flows just not that important Plain vanilla approach: conventional techniques, models, metrics
Use capital flow panel dis-aggregated by direction/type, 1990Q1-2015Q4, 85 countries
Measure GFCy via: a) observables in central countries (VIX) and b) unobservables (commonality in capital flows, via extracted dynamic factors)
Little evidence of significant, conspicuous GFCy
Using both measures gives upper bound of ≈25% of capital flow variation explained by GFCy
Often much less Rose: Quantifying Global Financial Cycle on Capital Flows 3<br>
slide4. Much Interest in Global Financial Cycle (GFCy) Literature
Rey and co-authors (many)
Forbes and Warnock (2012)
Ghosh et al (2014)
Koepke (2015)
Bruno and Shin (2015a, 2015b)
Avdjviev et al (2016 a,b)
IMF 2017 Annual Research Conference
RIDGE 2017 Conference Rose: Quantifying Global Financial Cycle on Capital Flows 4<br>
slide5. Serious Policy Implications Very synchronized GFCy implies EM policymakers can do little more than insulate their economies (capital controls, macro-prudential, ...) and blame US for exogenous shocks:
“As capital flows respond to US monetary policy, they may not be appropriate for the cyclical conditions of many economies...”
- Rey (2015, pp 9-10) Rose: Quantifying Global Financial Cycle on Capital Flows 5<br>
slide6. Identifying GFCy: Empirical Strategy GFCy intrinsically unobservable
Guiding idea: if GFCy consequential, should drive high proportion of fluctuations in most types of capital flows (and domestic credit creation, asset prices, ...) in many places, much of the time
High commonality in financial conditions, manifest in capital flows, driven by observable global determinants
The initial motivation: if the GFCy is so obviously important, why didn’t we already know about it?
Plain vanilla approach: model, data, estimation, metrics …. Rose: Quantifying Global Financial Cycle on Capital Flows 6<br>
slide7. Two Approaches to Measuring GFCy (Direct) Examine observable “fundamental” GFCy drivers, center-country macro/financial determinants of capital flows
Which variables?
Which center countries?
Literature delivers consensus (VIX), but check with many variables, countries
(Indirect) Examine observable commonality via factors
Extract common dynamic factors (following Rey)
Attribute all commonality to GFCy to be conservative, establish upper bound for GFCy impact
Ignore correlated shocks bc commodity price movements, sunspots, contagion, etc. Rose: Quantifying Global Financial Cycle on Capital Flows 7<br>
slide8. Fundamentals: Volatility VIX as consensus
Ticker symbol for CBOE implied near-term volatility of S&P 500 index options (end of quarter)
Used extensively in literature: e.g., Passari and Rey (2015, p683) VIX as “our proxy for the global financial cycle ... Large gross cross-border flows are moving in tandem across countries regardless of the exchange rate regime, they tend to rise in periods of low volatility and risk aversion and decrease in periods of high volatility and risk aversion, as measured by the VIX…”
Also used by Cerutti, Claessens, and Ratnovski (2017), Obstfeld et al. (2017), Advjiev et al. (2016), Bruno and Shin (2015a, b), Fratzscher (2012), Forbes and Warnock (2012) ...
Robustness: use VDAX (Germany), VSTOXX (EMU), IVI (UK) Rose: Quantifying Global Financial Cycle on Capital Flows 8<br>
slide9. Fundamentals, continued Use seven other fundamentals for sensitivity analysis as well
Nominal policy interest rate (FedFunds/UK Base/Euro Area deposit)
Ex post real rate (using realized CPI inflation)
TED spread (3-m LIBOR – government rate)
Yield curve slope (10yr – 3m government rates)
GDP growth
Real effective (CPI) exchange rate growth (Shin)
M2 growth
Default center-economy is USA
Use UK, EMU for sensitivity analysis Rose: Quantifying Global Financial Cycle on Capital Flows 9<br>
slide10. Capital Flow Data: Also Conventional Conventional Balance of Payments Series, 1990Q1-2015Q4, 85 countries (with gaps)
Usually focus on 63 “small” countries
Exclude US, UK, actual/future EMU members, Japan
Two directions (in/out)
Gross inflows/outflow: net transactions by foreigners/domestic residents
Different types
Foreign Direct Investment (FDI)
Portfolio Equity
Portfolio Debt
Sometimes aggregate portfolio equity and debt as “Portfolio”
Bank Credit (subset of “other investment liabilities, of which borrowers are banks”) Rose: Quantifying Global Financial Cycle on Capital Flows 10<br>
slide11. Factors: Measuring Capital Flow Commonality Create 180 common factors to ensure insensitivity
2 time spans (start 1990Q1 and 1996Q1; end 2015Q4)
3 sets of countries (advanced/emerging/both)
2 directions (in/out)
4 major types (FDI/debt/equity/credit) (also: portfolio≡debt+equity)
3 estimators
Dynamic factor with 1 lag, extract largest eigenvalue factor (Miranda Agrippino-Rey)
Ditto, with 2 lags
Traditional static factor

Ex: use dynamic factor analysis to extract a factor from fifteen emerging markets over 1990Q1-2015Q4 for FDI inflows
Include this in FDI inflow equations below Rose: Quantifying Global Financial Cycle on Capital Flows 11<br>
slide12. (National) Capital Flow Equations Panel regressions in paper indicate national heterogeneity
So estimate disaggregate capital flow equations (over time) with LS:

CAPFLOWd,e,i,t = ΣβijUSFUNDjt + ΣγkFACd,ekt + φi + εd,e,i,t for country i

Regressand: capital flows (% GDP) for direction/type/country
GFCy Regressors:
[Direct] Eight fundamentals (VIX, nominal and real rates, …)
[Indirect] Two factors for relevant capital flow (advanced/emerging)
By Design: ignore dynamics, national demand/supply shocks
Deliberately attribute all fit to GFCy Rose: Quantifying Global Financial Cycle on Capital Flows 12<br>
slide13. (National) Capital Flow Equations, continued Rose: Quantifying Global Financial Cycle on Capital Flows 13<br>
slide14. National Capital Flow Time-Series Regressions Rose: Quantifying Global Financial Cycle on Capital Flows 14<br>
slide15. Zooming in to the top-left Rose: Quantifying Global Financial Cycle on Capital Flows 15<br>
slide16. Sensitivity Analysis Rose: Quantifying Global Financial Cycle on Capital Flows 16<br>
slide17. Box Plots: National Capital Flows Time-Series Rose: Quantifying Global Financial Cycle on Capital Flows 17<br>
slide18. Different Specifications Rose: Quantifying Global Financial Cycle on Capital Flows 18<br>
slide19. Different Samples Rose: Quantifying Global Financial Cycle on Capital Flows 19<br>
slide20. Cutting the Sample Further Rose: Quantifying Global Financial Cycle on Capital Flows 20<br>
slide21. Summary Focus here: quantifying importance of GFCy for capital flows
Coming soon: credit growth, asset prices
Panel regressions, national capital flow equations, event studies
Plain vanilla approach: data, statistical metrics, models, techniques
Measure GFCy both directly and indirectly
Up to eight fundamentals for three large economies, though focus on VIX
Also ... commonality, extracted from dynamic factor models
Conservative approach
All explanatory power attributed to GFCy
Ignore coefficient signs, sizes (usual focus) ... which are often bizarre (Table A3)
Poor fit; typically < 25% variation explained, often much less!
Can find exceptions, but they are … exceptions Rose: Quantifying Global Financial Cycle on Capital Flows 21<br>
slide22. Conclusion Global Financial Cycle simply not that important in understanding capital flows
Capital flows of peripheral countries explained by idiosyncratic phenomena
If foreign: difficult to put in place systematic policies to screen “good” from “bad” capital flows
If domestic (likely): periphery country needs to take ownership
Harder to blame center countries for volatility if does not stem from center
Caveat: GFCy can appear elsewhere (e.g. asset prices and credit)
Our future research Rose: Quantifying Global Financial Cycle on Capital Flows 22<br>
slide23. Auxiliary Material<br>
slide24. More Motivation “There is a global financial cycle in capital flows, asset prices, and in credit growth. This cycle co-moves with the VIX, a measure of uncertainty and risk aversion of the markets.”
- Rey (2013, abstract)
“Large gross cross-border flows are moving in tandem across countries regardless of the exchange rate regime, they tend to rise in periods of low volatility and risk aversion and decrease in periods of high volatility and risk aversion, as measured by the VIX … There is a global financial cycle.”
-Passari and Rey (2015, p 693) Rose: Quantifying Global Financial Cycle on Capital Flows 24<br>
slide25. External Influences and the Blame Game Consider something important, driven by exogenous external effects
Flight delays (≡ scheduled-actual arrival time)
Linear models (with LARGE data sets) of flight delays often include:
Flight distance
Airline (carrier) effects
Time Effects : Hour (both departure and arrival), Day, and Month
Controls for conditions at both destination and arrival airports:
Airport traffic (incoming/outgoing)
Visibility
Temperature
Precipitation
A typical such model has R2 << .2!
But airlines often blame such forces for delays Rose: Quantifying Global Financial Cycle on Capital Flows 25<br>
slide26. Country List Rose: Quantifying Global Financial Cycle on Capital Flows 26<br>
slide27. Quick Peek at Capital Flows and VIX: Little Rose: Quantifying Global Financial Cycle on Capital Flows 27<br>
slide28. (Interestingly) Dismal Factor Results If GFCy drives most variation in types/directions of capital flows, then 8 factors should be strongly related
If VIX highly correlated with GFCy, it should also be correlated with factors
BUT THEY’RE NOT! Consider Advanced Economy Factors
Most scatters are simply clouds
VIX not linked to factors
3/8 correlations negative; all < .5
3/8 correlation coefficients significantly different from zero, 2 >0, 1 < 0
FDI inflows the strongest
Insensitive results (robustness in paper) Rose: Quantifying Global Financial Cycle on Capital Flows 28<br>
slide29. Capital Flow Factors and the VIX Rose: Quantifying Global Financial Cycle on Capital Flows 29<br>
slide30. Factors of Advanced, Emerging Countries Typically very loose relationship
Again, exception is FDI inflows Rose: Quantifying Global Financial Cycle on Capital Flows 30<br>
slide31. Advanced and Emerging Economies Rose: Quantifying Global Financial Cycle on Capital Flows 31<br>
slide32. Price and Quantity Factors Rose: Quantifying Global Financial Cycle on Capital Flows 32<br>
slide33. Price Factors and Fear Measures Also loose relationships Rose: Quantifying Global Financial Cycle on Capital Flows 33<br>
slide34. Fear Measures and Asset Price Factors Rose: Quantifying Global Financial Cycle on Capital Flows 34<br>
slide35. Capital Flow Panel Regressions Pool observations across countries
Disaggregate by direction, type
Start without any time-varying regressors, just time fixed effects
Gives upper bound on all possible global phenomena (including GFCy)
Estimate:

CAPFLOWd,e,i,t = {φi} + {θt} + εd,e,i,t

Capital flows as % GDP, disaggregated by:
Direction (d) – in/out
Type (e) – FDI/equity/debt/credit
Country (i)
Quarter (t) Rose: Quantifying Global Financial Cycle on Capital Flows 35<br>
slide36. Impact of Common Time-FE for Fit, Country FE Rose: Quantifying Global Financial Cycle on Capital Flows 36<br>
slide37. Capital Flow Panel Regressions, results CAPFLOWd,e,i,t = {φi} + {θt} + εd,e,i,t

Fits poor for all directions/types of capital flow using within/overall R2
Insensitive to time sample/country size
Global phenomena unimportant if impact on all countries identical Rose: Quantifying Global Financial Cycle on Capital Flows 37<br>
slide38. But ... 2 big restrictions implicit No explicit modelling of any economic/financial shocks (e.g., GFCy)
Assumed homogeneity of responses of all countries capital flows to common shocks

Unpack these assumptions in two steps Rose: Quantifying Global Financial Cycle on Capital Flows 38<br>
slide39. Model GFCy directly and indirectly Remove time FE, replace with 8 standard American “push” regressors
a) VIX; b) real GDP growth rate; c) nominal policy rate; d) real policy rate; e) TED spread; f) yield curve slope; g) REER change; and h) M2 growth
Add dynamic factors (match direction/type of capital flow to regressand)
Attributes all commonality to GFCy
Two factors: advanced, emerging market
Include country FE
But maintain common response (across countries) to 10 regressors:

CAPFLOWd,e,i,t = ΣβjUSFUNDjt + ΣγkFACd,ekt + {φi} + εd,e,i,t Rose: Quantifying Global Financial Cycle on Capital Flows 39<br>
slide40. Impact of 8 Fundamentals for Fit: Identical Responses, Country FE Rose: Quantifying Global Financial Cycle on Capital Flows 40<br>
slide41. Center-Country Variables explain little variation in capital flows But above maintains assumption that responsiveness identical across countries
So, add country-specific effects of US variables (possibly at lag s) on capital flows:

CAPFLOWd,e,i,t = Σβij,sUSFUNDjt-s + ΣγkFACd,ekt + {φi} + εd,e,i,t Rose: Quantifying Global Financial Cycle on Capital Flows 41<br>
slide42. Country-Specific Responses to (up to 8) Fundamentals, Country FE Rose: Quantifying Global Financial Cycle on Capital Flows 42<br>
slide43. Better, but not Good VIX alone does little, even when lags included
Need 8 fundamentals (plus lags)
Still fits poor ... all R2s << .5
Profligate parameterization: best fit is within R2 .42, but only 5.8 (=2016/349) observations per coefficient, overall R2 of .08! Rose: Quantifying Global Financial Cycle on Capital Flows 43<br>
slide44. What About Extreme Events? Forbes and Warnock (2012) select extreme values of capital flows:
Capital Flight (increased outflow)
Retrenchment (decreased outflow)
Stop (decreased inflow)
Surge (increased inflow) – also studied by Ghosh et al (2014) Rose: Quantifying Global Financial Cycle on Capital Flows 44<br>
slide45. Forbes-Warnock events not explained by center-country phenomena Quasi R2 (probit/logit) or within R2 (LS). Panel estimates with country random/fixed effects as marked. 53 countries, 1990Q1-2009Q4. Rose: Quantifying Global Financial Cycle on Capital Flows 45<br>
slide46. Domestic Credit Growth: like Capital Flows Just a peek at preliminary results; more on credit growth and asset prices to come (in separate paper) Rose: Quantifying Global Financial Cycle on Capital Flows 46<br>
slide47. Domestic Credit Growth Rose: Quantifying Global Financial Cycle on Capital Flows 47<br>
slide48. Suppose GFCy only Relevant during Crises? Little suggestion in literature, but ...
Focus on events when VIX high
8 occasions when VIX closes above 30
6 year event study
Means with empirical (5,95) confidence intervals
Little evidence of significant shifts around crisis periods
Results insensitive to exact definition of crises (see paper) Rose: Quantifying Global Financial Cycle on Capital Flows 48<br>
slide49. Event Study (8 events) Rose: Quantifying Global Financial Cycle on Capital Flows 49<br>
slide50. Robustness: US$, not VIX (7 events) Rose: Quantifying Global Financial Cycle on Capital Flows 50<br>
slide51. Robustness: lower VIX threshold (19 events) Rose: Quantifying Global Financial Cycle on Capital Flows 51<br>
slide52. Robustness: changes, not levels (12 events) Rose: Quantifying Global Financial Cycle on Capital Flows 52<br>
slide53. FDI Inflows, other Crisis Events Rose: Quantifying Global Financial Cycle on Capital Flows 53<br>
slide54. Robustness: average, not close (7 events) Rose: Quantifying Global Financial Cycle on Capital Flows 54<br>
slide55. Extensions Out of sample techniques?
Non-linearities?
Adding intrinsic dynamics (long-run/short-run differences)?
Mostly: Improving capital flow equations by adding domestic demand/supply factors
Any domestic/GFCy correlation likely to reduce impact of latter
Includes endogenous domestic policy responses
These omissions a likely reason for poor fit of regressions Rose: Quantifying Global Financial Cycle on Capital Flows 55<br>