Forecasting and Policy Analysis System (FPAS)
Download
Report
Transcript Forecasting and Policy Analysis System (FPAS)
Nikhil Vellodi, Bank of Papua New Guinea
PFTAC Workshop, Apia, Samoa,
15th-23rd November, 2011,
“Improving Analytical Tools for Better Understanding”
Introduction
Part I: Theory
◦ Model Description.
◦ Solving and Estimating the Model.
Part II: Application to PNG
◦ The PNG Database.
◦ Impulse Response Functions and Historical Decompositions.
◦ Drawbacks, Things to Work on.
Conclusion
2
What is an FPAS model?
How and where are they used?
How do they relate to the current workshop objectives?
Introduction
3
Small macroeconomic model.
Built at the IMF: Berg, Karam and Laxton, 2006, “A Practical
Model-Based Approach to Monetary Policy Analysis: A How-To
Guide”, IMF Working Paper 06/81.
Similar to many workhorse models used at central banks.
Mainly differs in the estimation technique.
Blends New Keynesian theory with DSGE modeling.
◦ No explicit microfoundations, as in a DSGE model.
◦ Uses IS/LM approach to aggregate demand and supply
Price stickiness in the short-run and demand equation/price setting equation.
◦ Demand and price-setting equations taken from New Keynesian Theory.
Phillips Curve and Output Gap equations broadly derived from Calvo (1982)
price setting rules.
Introduction
4
Used for forecasting and monetary policy analysis.
Mainly short-medium term forecasting. Both qualitative
and quantitative analysis.
◦ Calibration and estimation, using actual data, rather than pure
calibration as for DSGE or CGE models.
Explicitly models the monetary transmission
mechanism.
Mainly used in central banks and other financial
economic institutions.
Introduction
5
Provide thorough yet concise analysis of current
macroeconomic conditions, and implications for policy
going forward.
Draw together key macroeconomic variables, such as
the monetary and fiscal policy stances, the real
exchange rate, the output gap, inflation and the real
interest rate, all of which will be discussed in detail
over the course of the workshop.
Introduction
6
The Model Description for PNG.
Solving the model.
Estimating the model.
Part I: Theory
7
Overview of Model Structure
◦ Global economy split into Home and Rest of World.
External conditions are a major determinant of domestic conditions,
especially for small, open economies.
◦ Six main equations govern each region:
Aggregate Demand equation (Output Gap equation).
Price-setting equation (Phillips Curve).
Exchange Rate equation (Uncovered Interest Parity condition).
Monetary policy reaction function (Taylor Rule).
Fiscal policy reaction function (Fiscal Balance equation).
Non-mineral revenue equation.
Part I: Theory: The Model Description
8
The Output Gap equation describes the determinants of
aggregate demand.
Output is determined by past and future output, as well as the
real interest rate, the real exchange rate, real commodity
prices, the non-mineral fiscal balance and world output.
Part I: Theory: The Model Description
9
The Phillips Curve describes the trade-off between prices and
output.
An increase in the output gap implies a build-up of demand
pressures, which lead to inflation.
Core inflation is determined by past and future core inflation, the
past output gap and lagged headline inflation.
Part I: Theory: The Model Description
10
The Interest Parity Condition links the real exchange rate with the real
interest rate.
An increase in the domestic real interest rate, ceteris paribus, leads to
capital inflows, and hence an appreciation of the real exchange rate.
The real exchange rate is determined by the future real exchange rate,
the difference between domestic and world interest rates and the
domestic risk premium on investment.
Part I: Theory: The Model Description
11
The Taylor Rule describes the manner in which monetary policy is
conducted.
In response to an increase in inflation, the central bank raises the
nominal interest rate, in order to reduce demand and hence
inflation.
The nominal interest rate is determined by the past rate, the real interest
rate, core inflation, future deviations of core inflation from trend and the
output gap.
Part I: Theory: The Model Description
12
The fiscal balance equation describes what determines the profile
of government spending and saving.
An increase in the output gap correlates with increased
government revenue, increases in commodity prices increases
revenue through export receipts and dividend payments.
The fiscal balance is determined by the past balance, past output
and past commodity prices.
Part I: Theory: The Model Description
13
The non-mineral revenue equation was inserted to allow a
richer description of the role of non-mineral revenue in the
model.
By separating the fiscal balance into both revenue and price effects, we can
directly shock the revenue term, rather than just the commodity price
term.
Part I: Theory: The Model Description
14
Simply links core and headline inflation.
Part I: Theory: The Model Description
15
In a couple of stages…
1. Find the “steady-state” solution.
Variables have settled on constant values in time.
To find SS, drop the t subscripts, since values same from one period to
next.
e.g. yt, yt+1, yt-1 all become y.
Not unique. Many SS solutions may exist.
2. Determine dynamic properties. “Two-point boundary” technique.
Pick two SS solutions. Then the dynamic solution is the path traced by
the variables in getting from one to the other.
Impulse response functions similar technique.
Start at one SS solution, perturb one shock variables, and see how the model
returns to equilibrium (may be same or different from starting point).
Part I: Theory: Solving the Model
16
Bayesian Estimation Techniques.
◦ Simply application of Bayes’ Rule:
◦ Start with a prior assumption on parameter distributions, before data
is analysed (Pr(M).
◦ Use likelihood function (Pr(D|M) )/Pr(D)) to generate posterior
(Pr(M|D)).
This is the impact of the data on the choice of the model.
◦ Modeling rationale for using Bayesian estimation.
Formation of priors goes hand-in-hand with economic understanding.
Choosing the initial values for the parameters requires an intuition for
their role in the economy.
Part I: Theory: Estimating the Model
17
The PNG Database.
Impulse Response Functions.
Historical Decompositions.
Drawbacks of FPAS model.
Things to Work on.
18
Non-mineral real GDP output gap (HP, 1600)
5.0
4.0
3.0
2.0
1.0
0.0
-1.0
-2.0
-3.0
The PNG Database
19
Output Gap and Core Inflation (Y-o-Y, Annual Rate, Right Axis)
5.0
25
4.0
20
3.0
15
2.0
1.0
10
0.0
5
-1.0
0
-2.0
-3.0
-5
Core CPI, Change y-o-y annual rate
The PNG Database
20
Mineral Revenue in % non-mineral GDP
25.00
20.00
15.00
10.00
5.00
0.00
The PNG Database
21
Deviation of Non-Mineral Deficit from MTFS Target
in % of Non-mineral GDP (+ Fiscal Expansion)
8.00
6.00
4.00
2.00
0.00
-2.00
-4.00
-6.00
-8.00
-10.00
Deviation of nonmineral revenues from long-run trend in % of nonmin GDP (+ expansionary fiscal policy/revenue shortfall relative to trend)
Deviation of fiscal expenditures from MTSF norm in % of nonmin GDP
Deviation of nonmineral fiscal deficit from MTFS target in % of nonmin GDP (+ expansionary fiscal policy)
The PNG Database
22
Nominal and Real Interest Rates (commercial lending rates,
weighted average total advances)
25
20
Commercial lending rates (weighted Average
Total Advances)
Real lending rate (core, y-o-y)
15
10
5
0
-5
-10
The PNG Database
23
Deviation of Real Effective Exchange Rate from Trend in % (+
Depreciation)
15
10
5
0
-5
-10
-15
-20
The PNG Database
24
Commodity Price Index (in US$ and in DC, real terms)
250
200
150
100
50
0
Commodity Price Index, 2005 = 100, includes both Fuel and Non-Fuel Price Indices
Real Commodity Price Index (Index expressed in Kina, divided by PNG Headline CPI), 2005=100
The PNG Database
25
Selected charts.
◦ Shocks to domestic demand and non-mineral .
◦ Charts generated from prior distributions.
Not using the posterior distributions generated via estimation.
Impulse Response Functions
26
Impulse Response Functions
27
Impulse Response Functions
28
For a given equation, demonstrates the relate explanatory
power of each variable throughout the sample period.
E.g. for the output gap, gives a reasonably clear and thorough
impression of what the model thinks is driving demand
conditions in the PNG economy.
29
30
The Chart tells a compelling story of PNG’s macroeconomic
history over the last decade…
◦ The global financial crisis clearly hit demand, through an appreciation
in the real exchange rate and external demand conditions, but a fiscal
spending stimulus compensated, keeping the output gap almost at
zero.
◦ In the years leading to that, there had been a period of fiscal
tightening, which had a negative effect on demand.
◦ The real interest rate goes through periods of being more or less
significant.
E.g. there was a large decline in real lending rates around the crisis (2008),
which had a stimulating effect on demand.
31
What it doesn’t talk about.
◦ Aggregate supply, current account.
These are captured indirectly through the other endogenous
variables.
Modeling changes in supply through potential output is difficult.
Too computational intensive, couldn’t estimate the model.
May require more explicit microfoundations, with sectors, etc.
Technical requirements
◦ Have to leave model estimation running overnight!
Sensitivity to priors
32
Elaborating the inflation process
◦ May split inflation into imported and domestic.
There have been secular shifts in the causes of inflation over the last
decade.
Refining the transmission mechanism.
◦ Spread between commercial lending rates and OMO rates large and
erratic.
Employing one interest rate to act as both the Bank signaling rate and the
domestic demand determining rate may be infeasible.
Could use two separate rates, and add an equation linking the two.
Revisiting the sample period.
◦ Was there a trend break in fiscal policy in the early 2000s?
33
How does the FPAS fit in with PNG and the other South Pacific
Islands?
◦ Compliments larger DSGE model, both in terms of time-frame and
explanatory power.
DSGE model has sectoral elaboration, models supply side in greater
depth.
Operates over the medium-long term.
Qualitative analysis only.
Calibration only, no data or estimation involved.
◦ Just building the model is useful in itself.
Compiling the database puts the user in touch with key data series.
Calibrating priors forces user to think about how key variables interact.
◦ Point towards which all of the current work is leading.
34
THANK
YOU!
35