DSGE Models for Macroeconomic Policy Analysis:
Solution Methods, Estimation and Analysis
Lawrence J. Christiano
We will begin by reviewing model solution methods.
This discussion will focus on the analysis of simple real business cycle
models. We will then turn to the modern workhorse in macroeconomics, the New
Keynesian model. We will discuss the basic properties and policy implications
of a very simple closed economy version of that model (a second week in the
course will be devoted to an open economy model). In recent years, much work
has been poured into the introduction of financial frictions into the New
Keynesian model. I will present a simple overview of some of that work.
Finally, if there is time we will discuss vector autoregressions and what it is
that we can infer about models from them.
Background readings:
handbook chapter; Journal of Economic Perspectives, interview and this.
Outline
1) Overview of tools for solving DSGE models: Perturbation methods (here is the MATLAB code that generated the last graphs in the handout, and here is a Dynare file that can also be used to solve the model in the handout).
a)
a much more detailed version of the handout,
including a discussion of projection methods.
b)
Here is an exercise, which works with a
version of the model in the handout that includes hours worked (as in the
standard rbc model). The third question shows how to build and simulate a model
in which there is nonstationary growth. The material is in this zip
file.
c) Background
readings: Christiano-Fisher (JECD, 2000), Ken Judd’s textbook.
2)
The simple New Keynesian (NK)
model without capital. We will build the model (almost) from its
foundations and describe its properties and implications for policy. Most of
the implications for policy will be ‘discovered’ in MATLAB exercises.
a)
First,
we (i) derive carefully the model’s equilibrium conditions; (ii) talk about the
classical dichotomy and how it does not occur when there are sticky prices;
(iii) discuss the apparent absence of `money’ from New Keynesian models; (iv)
define the natural equilibrium, a benchmark for policy analysis.
b)
Second,
we derive the log-linearized equilibrium conditions around a zero-inflation
steady state. (Though we will not do a detailed derivation of the linearized
Phillips curve, that is covered here.)
We will then do the series of Dynare
exercises described in NK_exercise.pdf, which accomplish three things (you can
find the pdf file, as well as the associated Dynare code, cgg.mod, here):
i)
Convey
basic intuition about the working of the New Keynesian model.
ii)
Show
how under ‘news’ shocks, inflation targeting might drive the interest rate in
the ‘wrong’ direction and inadvertently trigger an inefficient stock market
boom (Slides, manuscript; and section 3.2 of handbook chapter).
iii) This
Dynare code contains the seven non-linear equilibrium conditions of the Simple
New Keynesian model. It can be used to show how Dynare handles this case, and
to investigate the accuracy of the linearization strategy used in parts (i) and
(ii).
c)
Third,
we will use the model to discuss a `Fisherian’ scenario in which inflation and
the nominal interest rate move in the same direction and an `anti-Fisherian’
scenario in which the two variables move in opposite directions. We will
discuss how the Volcker disinflation in the US can be understood as a blend of
the two scenarios. (See also these notes.)
d)
Finally, we will use the model to discuss
three types of conditional forecasting situations. They allow one to answer
questions like `what will happen if we keep the interest rate high over the
next year’, or `what will happen to our economy if the world economy begins to
weaken?’
i)
How
to compute Odyssean forecasts. The code
for this can also be used for other things, such as quantifying the forward
guidance puzzle or studying the impact on the government spending multiplier
when the interest rate is held constant (say, because the effective zero lower
bound is binding). Application: characterizing `forward guidance puzzle’ (see
this code,
as well as the commentary at the start of the code)
e) How
to compute two more standard types of conditional forecast. The code in (i)
above can be used to compare all three types of conditional forecasts.
3)
Bayesian inference.
a)
Additional material on mixed
frequency observations and link between DSGE models and VARs, reading. Code for GDP
example in handout.
b)
Exercise. Code for estimation part of
exercise. Code for MCMC part
of exercise.
c)
Readings: Del Negro and Schorfheide,
Zellner’s classic 1971 text, Bernanke and Boivin on ‘Data Rich’
estimation.
d) Methods to do a
Bayesian version of Generalized Method of Moments and to construct more plausible
priors for DSGE models.
4)
Introduction to
financial frictions: frictions arising from problems originating in the non-financial
business sector (Christiano-Motto-Rostagno AER
2014) and frictions arising
from problems in the financial sector (Gertler-Kiyotaki AER2015). For
more detailed lectures on the GK material see this
and this.
5)
Identification and Vector Autoregressions.
a)
Exercise, code. Answers by
Sebastian Kohls.
b)
Readings: CEE handbook chapter,
ACEL.
c) Information
on material not covered in lectures: for high frequency identification, see
Gorodnichenko and Weber; Gertler and Karadi; Nakamura and Steinsson, For sign
restrictions see Uhlig; Arias, Rubio-Ramirez and Waggoner.