Chapters on Monetary Policy
Dissertation DefenseMauricio Villamizar
Advisor: Professor Guido Kuersteiner
Committee: Professor Behzad Diba and Professor Matthew Canzoneri
1 “Identifying the Effects of Simultaneous Monetary Policy Shocks”
Job Market Paper (2014)
2 “Great Expectations? Evidence from Colombia’s Exchange Rate Survey”
Working Paper No. 735 at the Central Bank of Colombia (2012)
3 “The Impact of FX Intervention in Colombia: An Event Study Approach”
Forthcoming in the Journal of Desarrollo y Sociedad (2014)
Other Papers
“The Effects of FX Intervention: Evidence from a Rule-Based Policy Discontinuity in Colombia”
“The Impact of Pre-Announced Day-to-Day Interventions on the Colombian Exchange Rate”
Chapter 1
“Identifying the Effects of Simultaneous Monetary Policy Shocks”
Main Findings
Interest rate interventions have a significant impact on the economy but FX interventions have almost no effect
Empirical anomalies can be eliminated when accounting for the systematic responses of policy
Motivation
Several economies who claim to have a floating exchange rate (under an inflation targeting regime) do not really float
Motivation
The impossible trinity (trilemma) indicates that a country cannot Allow for free capital flows
Have autonomous monetary policy Adopt a fixed or managed exchange rate
In the empirical literature, there is a lack of consensus regarding the effectiveness of Central Bank intervention
US: Angrist et al. (2004, 2009, 2011, 2013), Bernanke et al.(1998), Christiano et al. (1996, 1998), Romer and Romer (1989, 2004), Rudebusch et al. (1998)
Latin America: Echavarria et al. (2009a, 2009b, 2013), Dominguez et al. (2012), Adler and Tovar (2011), Toro et al. (2010, 2005), Kamil (2008)
Colombia
The Central Bank of Colombia has adopted a policy framework consisting of two policy instruments
Interest rate Interventions (IRI) Foreign Exchange Interventions (FXI)
Research Objective
1 Model monetary policy behavior (parametrically) in order to extract
policy shocks, while leaving the response of the economy unspecified
2 Estimate the effect of these shocks (non-parametrically) on
economic variables
I employ:
Proprietary data(Direct interventions and internal forecasts)
Challenges
Empirical estimation faces 3 main challanges
1 Measurement Error: Not being able to observe policymakers’ exact
decisions
2 Simultaneity bias: Central banks react to economic conditions, and
economic conditions react to central bank interventions
3 Omitted Variable bias: Not capturing the relevant variables to model
policy behavior
I address (1) and (2) with the information and high frequency of my data set
Figure: Different Mechanisms of FX Intervention: 1999-2012
Figure: Intervention Interest Rate: 1999-2012
Internal Forecasts of Central Bank
Exchange Rate Misalignment Forecasts(et−Forecast(et))
7 “in house” structural models based on PPP, SVEC methodologies, current account equilibrium and HP filters
Inflation Forecasts(Forecast(πt)−πTargett )
Monetary Transmission Mechanism model with 9 equations based on prices, aggregate demand, wages, UIP condition, risk premium, etc.
Long Term GDP Forecasts (yt−Forecast(yt))
Other Variables Considered
Net position of the CBoC
Total net credit/debit with respect to the financial system
Board Meetings
Analogous to meeting dates of the US Open Market Committee
Capital Controls
During 2007-2008 investors had to deposit 40% of the inflow at the Central Bank during 6 months without interest payments
Other Variables
Exchange rate Volatility International Reserves US Fed Funds Rate Industrial Production
Methodology
Overview
Model the undertakings of monetary authorities
Extract the unexpected component of policy (policy surprises)
Independent Policies
FXI Policy Function:
FXIt =max[0, x
0
1tβ1+vt]
vt∼N(0, σ12)
IRI Policy Function:
Independent Policies
FXI Shock:
1t = FXIt−E[FXIt |x1t]
= FXIt−
Z
FXIt>0
(FXIt)dF(FXIt|x1t)
= FXIt−Φ x10tβ1
σ1
! "
x10tβ1+σ1φ
x10tβ1
σ1
!
/Φ x
0
1tβ1
σ1
!#
IRI Shock:
2t =IRIt−x
0
Dependent Policies
There is no reasona priori to believe that instruments are independent
A bivariate process forFXIt andIRIt can be described as:
FXIt∗ = x10tβ1+vt FXIt = max[0, FXIt∗]
IRIt = x20tβ2+2t
vt 2t ∼N 0, σ2 1 σ12
σ12 σ22
Dependent Policies
Constructing the Maximum Likelihood Function (in 2 stages):
DefiningA≡σ21−
σ212 σ2 2
andb≡x10tβ1+σσ122 2
(IRIt−x
0 2tβ2)
Stage 1: When FXIt >0 (FXIt =FXIt∗)
f(FXIt,IRIt) = f (FXI
∗
t |IRIt,x1t,x2t)f (IRIt |x1t,x2t)
= 1
A1/2φ
FXIt∗−b A1/2
1
σ2φ
IRIt−x
0 2tβ2
σ2
Dependent Policies
Stage 2: When FXIt = 0 (FXIt∗≤0)
f (FXIt,IRIt) = Pr(FXIt∗≤0|IRIt,x1t,x2t)f(IRIt |x1t,x2t)
=
1−Φ
b A1/2
1
σ2φ
IRIt−x
0 2tβ2
σ2
!
Stage 1 + Stage 2
Ln(θ) =
Y
FXIt∗≤0
1−Φ
b
A1/2
Y
FXIt∗>0
1
A1/2φ
FXI∗
t −b
A1/2
" Y 1 σ2
φ IRIt−x
0
2tβ2
σ2
IRFs
IRFs were estimated using equations (12) and (13):
Romer and Romer (2004):
Yit=γ0+
h
X
j=0
γj1t−j+ h
X
k=0
γk2t−k +ςit (12)
Jorda (2005):
Yit+s =ηs0+η
s
Table: Tobit Estimation: FXIt =max[0,x10tβ1+vt] +1t
Var / Specification (x10tβ1) (1) (2) (3)
FXIt−1 0.51*** 0.36*** 0.35***
(0.058) (0.056) (0.058)
et−1−Forecast(¯et−1) -2.36** -4.63*** -6.69***
(1.017) (1.078) (1.410)
Forecast(πt−1)−Target(πt−1) -7.41 -4.45 -5.81
(7.965) (8.121) (8.045)
yt−1−Forecast(¯yt−1) -40.8*** -66.0*** -47.4***
(11.901) (12.322) (13.390)
DCapitalControls -164.8*** -164.6***
(19.140) (20.192)
∆Rest−1 -0.3
Table: OLS Estimation: ∆IRIt=x20tβ2+2t
Var / Specification (x20tβ2) (1) (2) (3) (4)
∆IRIt−1 0.36*** 0.19*
(0.093) (0.110)
IRIt−1 -0.06*** -0.07***
(0.015) (0.012)
et−1−Forecast(¯et−1) 0.00 0.01***
(0.006) (0.005)
Forecast(πt−1)−Target(πt−1) 0.07***
(0.023)
yt−1−Forecast(¯yt−1) 0.08*** 0.08***
(0.025) (0.015)
∆it1−year1 0.63*** 0.76***
Table: Covariances of Bivariate Process
Specification x1t(1) x1t(2) x1t(3)
x2t(1) -0.04 -0.04 -0.03
(0.058) (0.061) (0.128)
x2t(2) -0.02 -0.04 -0.09
(0.054) (0.058) (0.215)
x2t(3) -0.04 -0.03 -0.04
(0.058) (0.061) (0.125)
x2t(4) -0.03 -0.05 -0.11
(0.053) (0.055) (0.177)
FXIt = max[0, x
0
Figure: Observed Policy Instruments vs Policy Shocks: 1999-2012
Figure: Implied IRFs of Inflation
(
π
t−
target(
π
t))
−3 −1.5 0 1.5 3 %0 5 10 15 20 25
periods after shock
delta IRI CI upper CI lower
(a) Response to a 1% change in ∆IRI
−3 −1.5 0 1.5 3 %
0 5 10 15 20 25
periods after shock
OLS residual CI upper CI lower
Summary FXI Shocks (1t) IRI Shocks (2t)
Inflation 0 − (≥1 year)
Industrial Production 0 − (months: 10-12)
Aggregate Demand 0 − (months: 11-15)
Exchange Rate 0 − (≤15 days)
Exchange Rate Volatility − (≤16 days) + (≤11 days)
Conclusions
Empirical anomalies, such as the price puzzle, are eliminated when accounting for the systematic responses of policy
FXI are not effective for depreciating domestic currency but they do have a small effect on reducing exchange rate volatility
A 1% increase in the intervention interest rate raises the 1-year Treasury bond’s yield by up to 0.25%
Policy positively impacts different maturity rates
Conclusions
Chapter 2
Great Expectations? Evidence from Colombia’s Exchange rate Survey
Conclusions
Main Findings
Short term expectations outperform a random walk process and do not exhibit a risk premium
Conclusions
Motivation
Internal discussion within the board of directors
How accurate were exchange rate expectations? Should they be included in internal forecasts/models?
Most theoretical models agree that expectations play a central role in the determination of the exchange rate.
Conclusions
Motivation
Most models in the international finance literature assume:
Rational Expectations
No time-varying risk premium
Homogeneous expectations
We find that these assumptions do not hold for the Colombian case
Conclusions
Data
We use the Central Bank Expectations Survey (Oct 2003-Aug 2012)
Conducted monthly to banks, stockbrokers and pension funds
(Traders and Analysts)
Survey asks for1-month,end-of-year, and1-year ahead exchange rate
Unbalanced panel with 4,389 observations for each forecast
15 establishments answered survey in 85% of cases 41 establishments answered survey in 50% of cases
Conclusions
Conclusions
Table: Accuracy of 1-month and 1-year forecasts
Entity Median Direction ∆St Direction ∆St +/−50 pesos +/−50 pesos
1-month 1-year 1-month 1-year
Banks 15 66% 35% 64% 9%
Stockbrokers 19 65% 43% 61% 15%
Conclusions Stabilizing Expectations
Deriving the Risk Premium
Ftt+k−St = it−ρembit−i ∗
t (CIP)
Et[St+k]−St = it−ρembit−i ∗
t −rpt (UIP)
Ftt+k−St = Et[St+k]−St+rpt (CIP+UIP)
= (Et[St+k]−St+k) + (St+k−St) +rpt
Conclusions Stabilizing Expectations
Estimations
For Panel regressions, we considered:
Fixed and random effects
To control for characteristics that could be correlated with the probability of an agent’s participation in the survey
Seemingly Unrelated Regressions
Conclusions Stabilizing Expectations
Efficient Market Hypothesis (EMH)
EMH can fail due to 1) a risk premium or 2) failure of rational expectations
Et[Si,t+k]−St=β0+β1(Ftt+k−St) +it
Table: Existence of Risk Premium
Variable 1-month 1-year
β0 -.006*** .012*** (0.000) (0.003)
β1 1.05*** 0.63***
(0.038) (0.030)
t:β1= 1 2.34 139.6*** (0.126) (0.000)
Et[Si,t+k]−St=α0+α1(St+k−St) +νit
Table: Rational Expectations
Variable 1-month 1-year
α0 -.005*** .04***
(0.000) (0.003)
α1 0.26*** 0.07***
(0.009) (0.008)
Conclusions Stabilizing Expectations
Orthogonality Condition
Are Agents capturing the impact of news and fundamentals?
Table: Et[Si,t+k]−St+k=xt0β+ηit
Fundamental 1-month 1-year
Board Meetingst -.012 0.31*** (0.014) (0.049)
Intervention Interest Ratet 0.00 0.003 (0.002) (0.007)
Ftt+k−St -0.52*** 0.24*** (0.062) (0.066)
St−St−k -0.08*** 0.044*** (0.015) (0.018)
Conclusions Stabilizing Expectations
Stabilizing Expectations
“Anticipatory purchases of foreign exchange tend to hasten the anticipated fall in the exchange value... and the actual fall may set up expectations of a further fall”
-Nurske 1944
Expectations can be seen as a linear combination of the spot rateSt and some
variablext
Et[St+k] =βxt+ (1−β)St
Candidates forxt used in the literature:
Past rate St−k → extrapolative
Expected past rateEt−k[St] → adaptive
Conclusions Stabilizing Expectations
Stabilizing Expectations
Table: Stabilizing Expectations
Type of Expectation 1-month 1-year
Extrapolative
Et[Si,t+k]−St=β0+β1(St−St−k) +t β1= -0.03** β1= -0.14*** (0.013) (0.015) Adaptive
Et[Si,t+k]−St=α0+α1(St−Et−k[St]) +νt α1= -0.07*** α1= -0.17*** (0.015) (0.017) Regressive
Conclusions
Out-of-Sample Forecasts
We set forth 5 competing strategies against a random walk process
In-Sample: Oct 2003 - May 2005
Used Rolling regressions to re-estimate parameters every forecast period 1-period out-of-sample forecasts were computed for each strategy
Constructed Mean Squared Prediction Errors (MSPEs)
MSPE =
N−1
P
i=0
[(E[St+k+ˆi]−St+i)−(E[St+k+i]−St+i)]2
N
Constructed 2 MSPE statistics: MSPEmodels
MSPERW
and(MSPERW−MSPEmodel)
Conclusions
Out-of-Sample Forecasts
Table: (MSPERW−MSPEmodel)
Strategy 1-month 1-year
Extrapolative -0.0006 0.18*** (0.001) (0.042)
Adaptive -0.0004 0.20***
(0.001) (0.045)
Regressive 0.003*** 0.09***
(0.001) (0.030)
Ftt+k−St 0.003** 0.03** (0.002) (0.016)
Conclusions
Conclusions
Revaluations were generally followed by expectations of further revaluation in the short run, but by expectations of devaluations in the long run
Financial establishments behaved poorly in terms of forecasting the direction and the level of the exchange rate
The Efficient Market Hypothesis does not hold for the Colombian case
Failure of rational expectations (short and long run) and existence of a risk premium (long run)
Chapter 3
The Impact of Foreign Exchange Intervention in Colombia: An Event Study Approach
Main Findings
Rule-based FX interventions was the only successful mechanism
Motivation
To date, there is still great controversy as to which exchange rate model should be used when measuring the effects of policy
It is important to understand the effects of policy without imposing parametric assumptions
Limitations
Subjectivity when choosing the window size of events
Large windows over-smooth density of the underlying data structure Small bandwidths reduce bias but increase variance
Event Window
Pre (-) and Post (+) Events
2, 5, 10 and 15 day windows
Event (Cluster of USD Purchases/Sales) Begins when central bank intervenes
Ends when central bank stops interventions for 2, 5, 10 or 15 consecutive days
For Example
Day 1 2 3 4 5 6 7 8 9 10
Intervention X X X X
Success Criteria
4 criteria for a successful Intervention-Frankel (1994), Fatum et al. (2001), and Humpage (1996)
Criteria Pre-Event Event Post-Event
Direction USD Purchases
+
∆St> 0
Reversal
−
∆St< 0 USD Purchases
+
∆St> 0
Smoothing
−
∆St< 0 USD Purchases
+
∆St> −
∆St
Matching USD Purchases
+
∆St> −
Table: Direction Criteria (5-day event)
FX Intervention Favorable cases H0:p≤0.5
(p-value)
Discretionary (Spot market)
USD Purchases 6/11 (0.27)
Discretionary (Options)
USD Purchases 11/19 (0.18)
Rules-Based (Options)
USD Purchases 7/11 (0.11)
USD Sales 7/9 (0.02)**
Table: Reversal Criteria (5-day event)
FX Intervention Favorable cases H0:p≤0.5
(p-value)
Discretionary (Spot market)
USD Purchases 5/11 (0.50)
Discretionary (Options)
USD Purchases 6/19 (0.92)
Rules-Based (Options)
USD Purchases 7/11 (0.11)
USD Sales 7/9 (0.02)**
Table: Smoothing Criteria (5-day event)
FX Intervention Favorable cases H0:p≤0.5 H0:p≤0.8
(p-value) (p-value)
Discretionary (Spot market)
USD Purchases 8/11 (0.03)** (0.62)
Discretionary (Options)
USD Purchases 12/19 (0.08)* 1 (0.93)
Rules-Based (Options)
USD Purchases 10/11 (0.00)***1 (0.09)*
USD Sales 9/9 (0.00)***1 (0.00)***
Table: Matching Criteria (5-day event)
FX Intervention Average Difference H0: +
∆St>
− ∆St
(p-value)
Discretionary (Spot market)
USD Purchases 0.06 (0.42)
Discretionary (Options)
USD Purchases 0.05 (0.39)
Rules-Based (Options)
USD Purchases 1.08 (0.11)
USD Sales -0.72 (0.02)**
Conclusions
Rule-based options were successful according to all criteria
All intervention mechanism were successful according to the smoothing criterion
However, Brazil’s counterfactual exercise casts doubts on Discretionary options