Macroeconometrics

Government Spending between Active and Passive Monetary Policy

Joint work with Sebastian Laumer (UNCG)

Q: Does the government spending multiplier depend on whether the central bank is "hawkish" or "dovish" towards inflation?
A: Sort of. Government spending multipliers seem to be larger when the Fed is "dovish" or "passive" towards inflation and does not raise interest rates when inflation rises--but only if they stay that way for a long time (more than two years). This does not happen. The Zero Lower Bound is the only plausible exception, and it could be something else entirely. That being said, the passive regime from our pre-2008 model exactly fits the economic dynamics observed during the ZLB era.

Abstract:

Conventional wisdom suggests that the government spending multiplier is larger when nominal interest rates respond less than one-to-one to inflation. However, models supporting this consensus estimate multipliers while holding the monetary policy rule constant after a government spending shock. We show that the multiplier does not depend on the monetary policy rule. What we find is that the monetary policy rule itself changes after a government spending shock and converges quickly to a similar regime regardless of the initial condition. This rapid change in monetary policy leaves the multiplier unaffected by the initial monetary policy regime. An exception to this characterization of monetary policy occurs when nominal interest rates are stuck at zero. We analyze the multiplier at the zero-lower bound and find that the multiplier exceeds one.

Does the Government Spending Multiplier Depend on the Business Cycle?

Answer: No, at least not as evinced by the data. Different models estimate different multipliers, but none of them seem to depend on whether the economy is in an expansion or recession--except in the very short run.

Joint work with Sebastian Laumer (UNCG)

Abstract:

We investigate the state-dependency of the government spending multiplier across the business cycle using a nonlinear two-regime VAR model. We find little evidence that multipliers vary between expansionary and recessionary periods. This is because the state of the business cycle itself changes after government spending shocks and converges towards a similar state. This result holds true regardless of how we model the business cycle. Furthermore, we conduct a rigorous robustness analysis to show that assumptions about the economic state built into linear impulse response functions are the key driver of the state dependency reported in the literature.

Replication Code can be found here for our sampler and GIRFs. 

Comments on estimating Smooth-Transition Vector Autoregressions

An early draft of a companion note to two empirical macroeconomics papers with Sebastian Laumer. Here, I expound on the smooth-transition VAR model and some of the difficulties associated with its estimation. These difficulties can be overcome quite easily by changing the specification of the model. 

Abstract:

The smooth-transition Vector Autoregression proposed by Auerbach and Gorodnichenko suffers from difficulty in estimation because its posterior is nonstandard. We explore this problem in detail and discuss the samplers available for that model. Then, we propose a different model specication that we believe to be more appealing a priori. We find that our new model specification has a standard Normal-Inverse-Wishart posterior, meaning that Gibbs sampling and conjugate priors can be used. The change in parametrization is resolves all of the issues with the Auerbach and Gorodnichenko ST-VAR model as found in the literature. Conditional on other parameters, the linear coefficients for two regimes are orthogonalized and can be drawn independently from a Gaussian posterior. The covariance matrices have conditionally independent Inverse-Wishart posteriors and standard conjugate priors can be used without requiring special treatment of the sampler.

Modifying the Gaussian Kernel for Kernel Density Estimation with Censoring

Q: Hey! That's not a macro paper!

A: Look at the application on page 20. If the Federal Reserve followed the same Taylor rule in 2008 as they had (on average) in the handful of years prior, they would have lowered the Fed Funds Rate to roughly -8% or -9%. But the zero-lower-bound intervened! Instead, they had to rely on unconventional policy such as quantitative easing, which was not rolled out for some time. See this paper by Wu and Xia who estimate the "effective" effective Fed Funds rate as a function of bond market data. There was a lag of several years before the Fed's policy caught up with where it would have been, had the zero-lower-bound not been a barrier. 

Abstract:

Kernel density estimation methods may be inconsistent if the data feature censoring. Following the literature, we provide a variation of the kernel density estimate which is consistent. Applied to the censored Gaussian kernel, this method then nests two classic models for censored regression by maximum likelihood. Furthermore, the revised kernel yields the correct likelihood in two applications where the Tobit model would be misspecied. We give a classical treatment of consistency for the maximum likelihood estimator based on that fact and provide simulation evidence that it is superior. Then, we give a quantile autoregression demonstration of the kernel density approach using the new kernel.