Transcript Unconventional Monetary Policy Rules

Unconventional Monetary
policy rules
SEM Conference, OECD July 22-24, 2015
Luca Agnello, Vitor Castro, Gilles Dufrénot, Fredj
Jawadi, Ricardo M. Sousa
1
Presentation
1- Context & Motivation
II-Balance sheet adjustment and term
spreads
III-Nonlinear rules
IV- Conclusion
2
I. Context and motivations
3
I. Context & Motivation #1
◦ Conventional Monetary Policy (CMP) with implicit or explicit
Inflation Target (IT) aims to preserve price stability in the
medium-term through the control of short-term nominal
interest rates.
◦ However, policy rates are bounded from below or at near the zero
floor.
◦ After the global financial crisis (2008-2009), several rate cuts have
been done, but showed the lack of effectiveness of this policy
during financial crises.
◦ Accordingly, several CBs switched after 2008 to Unconventional
MPR through Quantitative Easing (QE), and Large Scale Asset
Purchase (LSAP) Programs in order to affect long-term yields,
provide accommodative financial conditions and stimulate the real
economic activity (Chung et al., 2011).
4
I. Context & Motivation #2
◦ After 2008, the Fed expanded its balance sheet by about 1 trillion
US $ (7% of GDP). In 2009, the BoE set up an asset purchase facility
of about $275 billion. In 2009, the ECB initiates purchases of €60
billion and an asset purchase program of $61.5 billion was set up
by the BoJ in 2010.
◦ Three channels to influence long-Term Interest Rates (LTIR): i)
When CBs hold Short-Term Interest Rate (STIR) low for a long
period, ii) While helping to reduce the spreads between LTIRs and
STIRs, iii) Reducing the overall supply of long-term securities to push
up prices and reduce yields.
◦ Accordingly, UMP ⇒ More stabilization of Financial Markets and a
reduction of yields on targeted assets.
↪ Two Objectives: Monetary Stability & Financial Stability.
5
I. Context & Motivation #3
◦ The literature identifies a wide public debate on UMP (QE and
Credit Easing).
◦ The effectiveness of UMP is not unanimous (Eggertsson, 2006;
Curdia and Woodford, 2010): Why?
◦ While QE → ↓ LITR yields and term spreads→ Improvement of
financial markets (Meier, 2009; Gagnon et al., 2010),
◦ Its effect on real economy is not obvious: Gilchrist et al. (2009)
and Baumeister and Benati (2010) found a negative relationship
between the term-spread and the real economy; Chung et al. (2011)
and Perrsman (2011) showed that UMP is effective; the effect of
UMP is limited for Schenkelberg and Watzka (2013) and not
significant for Keister and McAndrews (2009).
◦ Are UMP rules misspecified? subject of Measurement errors?
6
I. Context & Motivation #4
◦ To better understand the transmission mechanisms of UMP, this
paper focuses on the UMP Rules with a special focus on two
instruments: i) the Term Spread (TS) and ii) the Growth Rate of
Central Bank Reserves (GRCBR).
◦ The interest of this paper is associated with the actual debate
concerning Conventional MPR and UMPR.
◦ The investigation of UMPR since 2008 enables to empirically
evaluate the UMP in the post-crisis context.
7
I. Context & Motivation #5
◦ The present paper contributes methodologically and empirically:
◦ Unlike previous studies, different flexible rule specifications are
estimated for different proxies for the UMP in order to apprehend
adequately the main effects associated with the conduct of UMP.
◦ The introduction of nonlinearity and asymmetry (through
different specifications) enables to capture further instability in this
rule. These flexibility and nonlinearity can capture the effects of the
UMP that was undertaken under different programs: QE1, QE2, QE3,
◦ Interestingly, the estimation of On/Off UMPR enables to estimate
nonstandard MP reaction functions and to identify regimedependence in the response of the TS to UMP.
8
I. Context & Motivation #6
We show that:
1. Neither stock prices nor housing prices affect, directly, the
conduct of UMPR, while UMP reaction functions are driven by the
dynamic of inflation and output-gap.
2. The presence of threshold effects in the UMPR, indicating that the
responsiveness of the term-spread to UMP is regime-dependent.
◦ For example, asset purchases during QE1 and QE3 have less
impact on the yield spread than in the remaining period of our
sample.
9
II. Balance sheet adjustment
and term spread
10
The estimated equations
 Equation 1: The central bank adjusts its balance-sheet in response to
 Macroeconomic environment (output-gap and inflation)
 Financial market developments (equity prices + housing prices)
 Central bank balance sheet: growth rate of central bank reserves (this is the
counterpart of the increase in the CB’s purchases of assets)
RG t  c   INFL INFLt   OG OGt   TS TS t   SPG SPGt   HPG HPGt 

k
 INFL
i k
t i

k
 OG
i k
t i

k
 RG
i k
t i

k
 SPG
i k
t i

k
 HPG
i k
t i
t,
RG : growth rate of central bank reserves, INFL : inflation rate, OG: output gap
TS: term spread (LTR-STR), SPG : changes in stock price, HPG: changes in house prices
11
The estimated equations
 Equation 1: The central bank adjusts its balance-sheet in response
to
RG t  c   INFL INFLt   OG OGt   TS TS t   SPG SPGt   HPG HPGt 

k
 INFL
i k
t i

k
 OG
i k
t i

k
 RG
i k
t i

k
 SPG
i k
t i

k
 HPG
i k
t i
t,
 Theoretical foundations for this rule : Woodford (2003), Goodfriend (2009),
Curdia and Woodford (2010),
-
Curdia and Woodford (2010), The central bank balance sheet as an instrument of
monetary policy, Carnegie-Rochester Conference on public policy “the future of
central banking”, JME, 2011,
↪ Such a rule is a “reserve-supply policy” or a “base money monetary policy”: the CB
chooses the quantity of reserves consistent with the bank’s target for short-term and long
term market rates,
↪ In addition, the optimal rule must take into account a representative consumer’s loss
function that depends upon the inflation gap, the output-gap and variables impacting the
consumer financial wealth (like term spreads in Woodford 2003). Here we consider
housing prices and equity prices because TS incorporates several channels of UMP.
12
The estimated equations
 Equation 1I: How the UMP has affected the term spread, given the
macro-financial environment?
TS t  c   INFL INFLt   OG OGt   RG RG t   SPG SPGt   HPG HPGt 
k
k
k
k
k
ik
ik
ik
i k
i k
  INFLt  i   OG t  i   RG t  i   SPGt  i   HPG t  i   t ,
RG : growth rate of central bank reserves, INFL : inflation rate, OG: output gap
TS: term spread (LTR-STR), SPG : changes in stock price, HPG: changes in house prices
 This equation can be interpreted as follows:
•
•
either as a causal relationship from linking CB’s reserves and term spread
or as a policy rule
13
The estimated equations
 Equation 1I: How UMP has affected the term spread, given the
macro-financial environment
TS t  c   INFL INFLt   OG OGt   RG RG t   SPG SPGt   HPG HPGt 
k
k
k
k
k
ik
ik
ik
i k
i k
  INFLt  i   OG t  i   RG t  i   SPGt  i   HPG t  i   t ,
 Policy rule :The main motivation is empirical:
 Can be interpreted as a “modified Augmented Taylor rule” obtained by adopting the
approach suggested by Razzak (2001)
 Based on the empirical observation that the relationship between RG and TS is
“stable”
 stability : cointegration over the period under study or stable coefficient (may be
problematic if not.)
14
Estimation methods and data
 Estimation methods
↪ DOLS + different IV estimators to account for endogeneity (2SLS, GMM)
↪ Robust standard errors (autocorrelation and heteroscedasticity)
 Data (source BOG of the Fed, BLS)

monthly frequency data for the United States over the period 2008:112014:12 (beginning of unconventional monetary policy)

Term spread: long-term (10-year) government bond yield rate - the shortterm (3-month) Treasury Bill rate

central bank reserves: data for the reserves of depository institutions

Consumer Price Index (CPI) - All Urban Consumers (all items less food
and energy, 1982-84=100) from the Bureau of Labor Statistics (BLS)

output gap : industrial production (HP filter)

Stock price : S&P 500 index

Housing price index: median sales price for new houses sold in the United
States (Federal Reserve Bank of St. Louis)
15
Main results

16
Main results
No evidence of forward guidance channel : Lower inflation CB purchases large
amount of assets signal a commitment to maintain low STR  expectations of low
STR and hence of low LTR  should narrow the term spread (positive sign)
No evidence of Financial market channel (portfolio re-balancing effect). By purchasing
large amounts of assets  CB reduces the relative supply of some assets increase in
their prices and lower their returns should narrow the term spread
17
Main results
Table 2. Unconventional monetary policy rules - reserves growth regressions, 2008:11-2015:01
(DOLS model).
VARIABLES
Inflation
OutputGap
Term Spread
Stock Price
Housing Price
Constant
Observations
R-squared
(1)
45.3027***
[3.230]
0.2653
[0.608]
3.6784**
[2.638]
0.3326
[0.715]
-0.0862
[-0.170]
-14.2228***
[-3.098]
71
0.428
Notes: Robust t-statistics in brackets. *** p<0.01, ** p<0.05, * p<0.1.

Key driver of reserve policy seems to be inflation (coefficient of high
magnitude)

Does it mean : no consideration for the business cycle or equity and housing
markets at all?
 Be careful : TS are statistically significant. There may be an issue of colinearity
here (TS: advanced indicators of the real activity and affects investors’ riskaversion in the equity markets).
18
Main results
Table 3. Unconventional monetary policy rules - term spread regressions, 2008:11-2015:01
(IV models).
Variables
Inflation
Output Gap
Central Bank Reserves
Stock Price
Housing Price
Constant
Observations
R2
Endog test
Overid test
Weak id test
S-Y 5% c.v.
(1)
iv/2sls
-4.9463***
(1.5533)
-0.1107***
(0.0311)
0.0071
(0.0084)
-0.0073
(0.0169)
-0.0113
(0.0155)
3.1474***
(0.2295)
74
0.0954
15.443
[0.0015]
20.680
[0.1474]
19.490
{19.29}
(2)
iv/2sls rob
-4.9463***
(1.4143)
-0.1107***
(0.0224)
0.0071
(0.0067)
-0.0073
(0.0127)
-0.0113
(0.0161)
3.1474***
(0.1981)
74
0.0954
8.311
[0.0400]
24.384
[0.0588]
19.490
{19.29}
(3)
iv/2sls hac
-4.9463***
(1.4422)
-0.1107***
(0.0281)
0.0071
(0.0072)
-0.0073
(0.0133)
-0.0113
(0.0144)
3.1474***
(0.1992)
74
0.0954
3.603
[0.3077]
16.951
[0.3218]
19.490
{19.29}
(4)
iv/gmm hac
-5.1691***
(0.9163)
-0.1028***
(0.0191)
0.0061**
(0.0025)
-0.0102
(0.0074)
0.0009
(0.0113)
3.1681***
(0.1400)
74
0.0635
3.603
[0.3077]
16.951
[0.3218]
19.490
{19.29}
Correcting for endogeneity :

The results remain unchanged for the term spread equation

CB rule : endogeneity is rejected for the RHS variables
19
Nonlinear rules
20
Some additional questions
 UMP were undertaken under different programs:
QE1, QE2, QE3, operation twist, ZLB, forward guidance,
 Can we find differences regarding their impact on the term
spread across time?
 At different stages of UMP, the goals have been
multiple
 To keep control on the short-term market rates by controlling
the demand for reserves by banks  interest rates on reserves
and TDF until 2009.
 To solve a liquidity crisis in the short-term monetary markets
 To affect the market rates and reduce the slope of the term
structure for the purpose of boosting the economy (in 2009 the
unemployment was around 10%.
 To « rebuilt » the housing, mortgage markets and other asset
markets
21
Some additional questions
 At different stages of UMP the goals have multiple
 These multiple goals can be seen as conditioning the strenght of
the Fed’s reaction to the variables in the RHS of both equations.
 They may also influence the sign of the coefficients if some of
the goals are « substitutes ».
 To study these questions, we need an empirical
framework which is more flexible
 To allow regime-switching dynamics in the term spread and the
CB’s reserves,
 To allow time-varying coefficients across time.
We present here several examples,
22
Example 1: was the output-gap, a
discriminating factor of term spread policy?
 TAR model
TS t   ' 0  '1 INFLt   ' 2 OG t   ' 3 RG t   ' 4 SPGt   ' 5 HPGt   1,t if S t  c
TS t   ' ' 0  ' '1 INFLt   ' ' 2 OG t   ' ' 3 RG t   ' ' 4 SPGt   ' ' 5 HPGt   2,t if S t  c
Transition variable (St), threshold parameter (c)
 Application linearity tests, for different candidate variables (lagged
TS, contemporaneous and lagged RHS)
 Linearity is highly rejected when the output gap is the transition variable
 A break date is identified corresponding to 2011:7
 In the term spread rule, inflation was not the key objective of UMP,
23
Example 1: was the output-gap, a
discriminating factor of term spread policy?
Table 3. Specific threshold tests.
Equation
TS
St
Tsay (1989) test a
Hansen (1996) test b
Threshold date c
Model
OG
0.0970
0.0060
2011:07
TAR(2)
 2011:7 as break date
 Change in unconventional monetary policy : end of QE2 (June 2011 and
beginning of Operation Twist (September 2011)
 The story to test
 Assume that the business cycle has been a guide for the setting of UMP
 Did this implied a higher willingness to fight output deviations (higher
coefficient)?
 Does it mean more “tolerance” for inflation (no significant response) ?
 Did this implied looking more carefully at other advanced indicators of the
activity (for instance the housing market and/or the equity markets).
24
Example 1: was the output-gap, a
discriminating factor of term spread policy?
Table 4. TAR estimation.
Constant
INFL
OG
RG
SPG
HPG
adjusted-R2
Log Likelihood
Durbin-Watson Statistic
Constant
INFL
OG
RG
SPG
HPG
adjusted-R2
Log Likelihood
Regime 1
-0.0950
[-1.1518]
0.7406
[1.2991]
-0.0145
[-1.1367]
-0.0178***
[-2.9379]
0.0193**
[2.2059]
-0.0017
[-0.2136]
0.27605
4.9595
1.4063
Regime 2
0.0061
[0.0939]
-0.1145
[-0.2630]
-0.1095***
[-3.1246]
0.0020
[0.2571]
0.0238***
[2.6355]
-0.0073
[-1.3218]
0.2242
19.1437
Regimes 1 and 2 : before and post
2011:7
Evidence of a switching TS rule
Output-gap significant in the second regime with a
value near the estimation obtained in non-switching
models
No evidence that inflation mattered (even after 2011:7)
This contradicts the view that operation twist was
undertaken to avoid inflationary pressure (QE: changing
the composition of assets+ credit easing)
Strong evidence that, over the whole period attention
has been paid to the development in the equity
markets
25
Example 2: QE1, QE2, QE3 : equally
efficient?
 Did the QE policies affect the term spread? No consensus in the
literature

Yes : LSAPs have reduced the slope of the term structure of the interest
rates (see, D’Amico, English, Lopez-Salido, & Nelson, 2012; Gagnon, Raskin,
Remache, & Sack, 2011; Krishnamurthy & Vissing-Jorgensen, 2011).

No : the positive effects of QE on market rates effect died out quickly (see
Chen, Curdia, & Ferrero, 2012; Wright, 2011)


Here
We observe that there was a significant relationship between TS and
CB’s reserves  need to see whether the magnitude has changed or
not across time,
26
Example 2: QE1, QE2, QE3 : efficient?

𝑇𝑆𝑡 =
𝜓1𝑖 OGt−i +
𝑖
𝜓2𝑖 INFLt−i +
𝑖
𝜓3𝑖 SPGt−i +
𝑖
𝜓4𝑖 HPGt−i +
𝑖
𝛾𝑖 𝑠𝑡 RGt−i + 𝜀𝑡
𝑖
27
Table 5. Unconventional monetary policy rules - term spread regressions, 2008:11-2015:01
(MS model).
Non-switching parameters
0.0384
[1.0620]
Output Gap (lag 2)
-0.2617***
[-6.7616]
Inflation (lag 1)
-0.0569
[-0.1515]
Inflation (lag 2)
-0.7760
[-0.8529]
Stock Price (lag 1)
-0.01463**
[-2.5165]
Stock Price (lag 2)
0.0067
[1.24405]
Housing Price (lag 1)
0.01858*
[1.7901]
Housing Price (lag 2)
0.0217**
[2.4230]
Switching parameters
Regime 1 (S=1)
Central Bank Reserves (lag 1)
0.0102***
[4.7405]
Central Bank Reserves (lag 2)
-0.0000
[-0.0022]
Constant
1.9239***
[9.3539]
0.0120
1
[1.15]
Regime 2 (S=2)
Central Bank Reserves (lag 1)
0.0479***
[2.7245]
Central Bank Reserves (lag 2)
0.0124
[0.8230]
Constant
2.6925***
[17.0935]
0.1176***
2
[2.71]
Notes: Robust t-statistics in brackets. *** p<0.01, ** p<0.05, * p<0.1.
Output Gap (lag 1)
The TS was less responsive to changes
in asset purchases in regime 1 than in
regime 2
The estimated coefficient is fourth as
high in one regime (regime 2)
28
 It seems that QE1 and QE3 programs, occurred during regime one,
exerted a moderate impact on the term spread in comparison to the
QE2 and the QE4 programs
29
Example 3: How did the Fed respond to asset
prices to improve economic outcomes?
 UMP have prompted controversy about this question, especially
when asset price bubbles and burst are likely
 No consensus about how changes in the equity and housing
markets have impacted the conduct of QE policies.
 View 1 : Asset price changes were early warning indicators of
inflation and output-changes (through the channel of wealth, cost
of capital, financial accelerator)
 There were no need for MP to directly target changes in asset
prices (or to the TS)
 MP paid attention to the rises and fall in financial prices, or to the
TS as far as they contained information about past inflation and
output
30
Example 3: How did the Fed respond to asset
prices to improve economic outcomes?
 In the CB rule, we terms spread or changes in the equity and
house price do not matter as targeted variables of UMP, but as
conditioning variables.
These variables are targets for macro-prudential policies (not UMP)
 View II: Since there were broad swings in asset prices, MP
searched to reduce financial market volatility, in addition to
controlling inflation and output
Necessary because we do not know how asset prices are related
to the economy (no consensus : double causality with inflation and
output? )
Caveat : possible contradiction (several objectives for only one
instrument)
31
Example 3: How did the Fed respond to asset
prices to improve economic outcomes?
 Theoretical models cannot be of clear guidance to distinguish
between these two views
Most models would be either normative (based on the optimal
monetary policy) or based simulations to compare different rules
(Models à la Filardo (2001), Bernanke and Gertler (2001), Cecchetti
et al. (2000))

We use an empirical test based on TVPMS models
32


 Table 6. Results of the TVPMS model
Variable
Coefficient
Std. Error
z-Statistic
Prob.
1.497748
0.051842
6.406156
1.489835
16.93679
0.1342
0.9587
0.0000
0.1363
0.0000
-1.319173
1.834880
-1.396396
1.311348
12.66597
0.1871
0.0665
0.1626
0.1897
0.0000
Regime 1
CONSTANT
INFLATION
OUTPUT GAP
TERM_SPREAD
LOG(SIGMA)
39.89502
3.421167
13.26575
16.07496
2.773482
26.63667
65.99200
2.070781
10.78976
0.163755
Regime 2
CONSTANT
INFLATION
OUTPUT GAP
TERM_SPREAD
LOG(SIGMA)
-3.753001
13.42590
-0.574811
1.444681
1.274190
2.844965
7.317043
0.411639
1.101676
0.100599
Transition Matrix Parameters
REGIME 1
SP GROWTH(-1)
HP GROWTH(-1)
REGIME2
SP GROWTH(-1)
HP GROWTH(-1)
-0.201614
0.193912
0.264215
0.189984
-0.763070
1.020678
0.4454
0.3074
0.092067
-0.396890
0.201848
0.109837
0.456121
-3.613447
0.6483
0.0003
35
Conclusion
36
What were the determinants of the Fed’s UMP ?
 First result obtained from RG and TS (linear models) : the Fed has
not targeted the house and stock prices directly, but focused on its
traditional targets, in particular inflation.
Two possible explanations
 One explanation : the financial environment has conditioned the
strength of the CB’s reaction to its traditional targets, while not
being directly targeted (results of the TVTPMS model)
 Another explanation : Specification error when we use a simple
rule augmented with the HP and SP variables
Second result : Different effects of QE policies on TS across time,
with a more moderate impact of QE1 and QE3
Third result : more tolerance to inflation has resulted in a higher
emphasis put on the output-gap (which contrast with the emphasis on
inflation in non-crisis periods).
37