Uncertainty and Monetary Policy:
Theory and Practice
Central Bank of Argentina, 2007
Money and Banking Conference
Joseph Tracy, Federal Reserve Bank of NY
Monetary Policy under Uncertainty
� Alan Greenspan, Jackson Hole WY, 29 Aug 2003
� “Uncertainty is not just an important feature of the monetary policy landscape; it is the defining characteristic of that landscape. As a consequence, the conduct of monetary policy in the United Statesat its core involves crucial elements of risk management, a process that requires an understanding of the many sources of risk and uncertainty that policymakers face and the quantifying of those risks when possible. It also entails devising, in light of those risks, a strategy for policy directed a maximizing the probabilities of achieving over time our goal of price stability and maximum sustainable growth that we associate with it.”
Risk management approach to monetary
policy
� Objective is not to maximize the benefits
based the expected evolution of the economy
� Rather, to maximize the probability of
achieving the goals of the Central Bank
� Essential elements
� Understanding of risks to the forecast
� Consequences associated with each risk
� Focus on the distribution of possible outcomes for inflation and growth
Examples illustrating this approach
� Russia Default, Fall 1998� FOMC eased policy despite view that economy
was expanding at a satisfactory pace and would likely continue to do so even without easing
� Insurance against any “contagion”
� Deflation scare, Spring 2003� FOMC eased policy even though deflation was
not viewed as the most probable outcome given the existing policy stance
� Insurance against the downside risk
� Monetary policymakers must make decisions under uncertainty� Where has the economy been (interpretation,
measurement)
� Where is it now (interpretation, measurement)
� Where will it be in the future (interpretation, expectations, transmission mechanism, future shocks)
� Quantifying uncertainty and monetary policy is controversial, difficult and an area of active research
� Presentation will describe current methods used to quantify future uncertainty at FRBNY
Decisionmaking under Uncertainty
� Central Banks can ignore uncertainty if certainty equivalence (CE) holds
� Strong assumptions required: Policymaker has a quadratic loss function, linear economy with all aspects known except future value of Gaussian shocks
� Many Central Banks act and communicate as if certainty equivalence does not hold
� Policy actions implemented in non-CE environments require dealing with uncertainty
� Bayesian approach quantify uncertainty through forecast distributions
� Robust Control approach does not quantify uncertainty
When Uncertainty Matters
� Use structural economic models
� Use reduced form time series models
� Using quantified judgment, present a forecast
distribution that uses a wide range of formal and
informal information.
� Elements from all 3 approaches can (and should) be
combined
Before explaining our judgmental approach need to agree on
some terminology
Methods for Constructing Forecast Distributions
� No consensus method of describing balance of risks or uncertainty around the judgmental forecast
� Central Bank point forecasts often interpreted as the mode: most likely outcome
� Other measures of “central tendency” are the mean
(expected value) and median (50% above, 50% below)
� When the median is not equal to mean, information
content in stating "balance of risks“
� For now focus on mean versus mode/judgmental
forecast
– upside risk to judgmental forecast: mean > mode
– downside risk to judgmental forecast: mean < mode
Balance of Risks
� Quantify uncertainty and balance of risks with forecast origin T (eg 2007Q1) and horizon h (eg 4 quarters)
with 3 numbers:
� µT+h is BoE central forecast, mode of forecast distribution
� σT+h is a measure of outcome dispersion around forecast
� λ percent of the time values lower than mode value are expected to occur
� If λ = 0.5 ⇒ mean=mode: risks are balanced
� If λ > 0.5 ⇒ mean less than mode: downside risk
� If λ < 0.5 ⇒ mean greater than mode: upside risk
Bank of England Fan Charts
� Choose central forecast, dispersion and amount of
downside risk.
� If risks are balanced then generate realizations from
a distribution centered at central point forecast with
given dispersion
� µiT+h = µT+h + εi , where εi ~ N(0,σt+h)
� If risks are unbalanced
� λ percent of realizations given byµi
T+h = µT+h − | εi |
� (1− λ) percent of realizations given by
µiT+h = µT+h + εi
Implementing Bank of England Approach
� Allow for dynamically varying balance of risks
� Interpret the upside and downside risks as scenarios
� Place positive probability on a scenario associated with
judgmental forecast
� Now three scenarios: central bank forecast, downside scenario, upside scenario
� Uncertainty now varies, consider two extreme cases:
� 50% weight on upside scenario, 50% weight on downside scenario – maximum uncertainty
� 100% weight on central scenario – minimum uncertainty
� Same implications for mean but very different implications
for uncertainty and changes in central forecast
First Generalization of BoE Approach
� Choose central forecast, dispersion and dynamic weights
on central scenario ( λ0,T+h ) and downside ( λ1,T+h )
� At each horizon h
� λ0,T+h percent of realizations generated by
µiT+h = µT+h + εi, where εi ~ N(0,σt+h)
� λ1,T+h percent of realizations generated by
µiT+h = µT+h − | εi |
� ( 1 − λ0,T+h − λ1,T+h ) percent of realizations generated by
µiT+h = µT+h + | εi |
Implementing Generalization
� Usually risks are balanced at forecast origin
� Eventually probability on central scenario goes to 1
� Central scenario at longer horizons constrained
� mean=mode=median ( long-run balanced risks )
� expected inflation at implicit target ( CB expects to
achieve inflation objective )
� expected output gap at zero ( CB expects to achieve
growth objective )
� uncertainty around these values close to historical averages
Constraints on Forecast Distributions
� Replace upside and downside risk scenarios with general scenarios about inflation and output
� Integrate scenarios with the assessment of uncertainty
� No fixed number of scenarios – keep to most important
� Examples:
� Productivity Boom: draw values of output > mode, inflation < mode
� Productivity Slump: draw values of output < mode, inflation > mode
� Overheating: draw inflation >> mode; initial output > mode, then output < mode
� Over-Tightening: output << mode, inflation << mode
Further Generalization of BoE Approach
1.0
1.5
2.0
2.5
3.0
2006 2007 2008 2009
1.0
1.5
2.0
2.5
3.0
Alternative Scenarios of Core PCE Inflation
% Change – Year to Year
Source: MMS Function (FRBNY)
Productivity Boom
Overheating
Productivity Slump
Central Scenario
Over-Tightening
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
2006 2007 2008 2009
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
Alternative Scenarios of Real GDP Growth
Source: MMS Function (FRBNY)
Productivity Boom
Productivity Slump
Overheating
Central Scenario
Over-Tightening
% Change – Year to Year
� Retain the same restrictions on dynamic evolution:
� Risks are balanced, usually at forecast origin, with initial weight 1 on central scenario
� Choose an initial probability weight for each scenario
� Choose how long each scenario will last if it occurs
� As forecast horizon increases, probability weight on central scenario goes to 1
� Generalize over earlier method by allowing for uncertainty over probabilities and duration of scenario
Probability Weights on Scenarios
Scenario Probabilities
Source: MMS Function (FRBNY)
0
10
20
30
40
50
Productivity Boom
Productivity Slump
Effects ofOverheating
Over-Tightening
0
10
20
30
40
50Percent
Prob. of being in the scenario in 2007Q4
Prob. of being in the scenario in 2008Q4
Prob. of being in the scenario in 2009Q4
Prob. of remaining in scenario through 2009Q4
� Plausibility of results based on introspection
� Does "continuing expansion" probability look sensible?
� Do inflation risks look sensible?
� Comparison with other measures
� Time series models� No forward judgment but allows for structural change in
past
Reliability Checks
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
2006 2007 2008 2009
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
Core PCE Inflation Forecast Distribution
% Change – Year to Year
Note: The probability interval shows the 50, 75, and 90 percent chance that the four quarter change in core PCE inflation will be within the respective range. The yellow line represents the expected value of the forecast, while the red line represents the actual FRBNY forecast.
Source: MMS Function (FRBNY)
-1
0
1
2
3
4
5
2006 2007 2008 2009
-1
0
1
2
3
4
5
Real GDP Growth Forecast Distribution
% Change – Year to Year
Note: The probability interval shows the 50, 75, and 90 percent chance that the four quarter change in real GDP growth will be within the respective range. The yellow line represents the expected value of the forecast, while the red line represents the actual FRBNY forecast.
Source: MMS Function (FRBNY)
Forecast
� Assume central bank uses a Taylor-type rule to determine FFR
from inflation and output gaps
� In the short run, adjust to discreteness of FOMC moves
� Choice of a particular rule and forecast distribution for inflation
and output produces a forecast distribution for FFR
� Examine variants on baseline policy rule
� Opportunistic Disinflation: slower rate cuts than baseline rule if inflation above 2%
� Dove: faster rate cuts than baseline if negative output gap
Distributions for Fed Funds Rate (FFR)
� Use futures and options prices to generate a forecast
distribution for the FFR
� Market distribution affected by
• Compensation for risk
• Market misunderstanding FOMC communication on objectives,
reaction function or outlook
• Markets view of the outlook and risks on real activity and inflation
� FRBNY distribution has been close to markets
� Sensitive to weight on productivity scenarios
� Another reliability check
� Do changes in market distribution and in our distribution line up?
Comparison to Market Forecast Distribution
3.0
3.5
4.0
4.5
5.0
5.5
2006 2007 2008 2009
3.0
3.5
4.0
4.5
5.0
5.5
Nominal FFR under Different Policy Rules
Percent
Source: MMS Function (FRBNY)
Baseline
Opportunistic Disinflation
Dove
Market-Implied
Comparison with BoE and BoG� Bank central forecast
� FRBNY judgment but at further horizons converges to assumed inflation target and zero output gap
� BoG judgment, FRBUS, etc.� BoE produced by numerous economists, MPC members and a
large number of models
� Dispersion
� FRBNY forecast error behavior, implied volatility and introspection
� BoG given by last 18-20 years of forecast errors for GB or residuals from FRBUS
� BoE exponential smoother of last 10 years of observed forecast errors
� Balance of Risks around forecast� FRBNY as described above� BoG does not allow explicitly for unbalanced risks� BoE produced by the MPC
Comparison with BoE and BoG
� Scenarios
� FRBNY scenarios produce forecast distribution
� BoG scenarios unrelated to forecast distribution
� BoE no explicit scenario analysis
� Option price information
� FRBNY forecast distribution converted to FFR using policy rule, dispersion calibrated to be similar to markets
� BoG forecast distribution of FFR from estimated policy rule in FRBUS compared to implied volatility
� BoE present implied volatility on short-term interest rate only
Role of scenarios and information from option prices