HANK economies with countercyclical risk
In this paper, Sushant Acharya and Jess Benhabib explore the dynamics of Heterogeneous-Agent New-Keynesian (HANK) economies, particularly focusing on the implications of countercyclical risk on monetary policy and equilibrium determinacy. Key Findings:
- Endogenous Natural Interest Rate:
In HANK economies with countercyclical risk, the natural interest rate is endogenous and co-moves with output. This makes the economy susceptible to self-fulfilling fluctuations, unlike in Representative-Agent New-Keynesian (RANK) models where the Taylor principle ensures a unique equilibrium. - Insufficiency of the Taylor Principle:
The Taylor principle, which suggests raising nominal interest rates more than one-for-one in response to inflation, is not sufficient to guarantee a unique equilibrium in HANK economies if risk is even mildly countercyclical. Multiple bounded equilibria can exist regardless of how strongly monetary policy responds to inflation. - Mechanism of Indeterminacy:
The paper explains that in a HANK economy with countercyclical risk, households face greater income risk during downturns, leading to increased precautionary savings and lower current spending. This dynamic can create self-fulfilling beliefs about economic downturns, acting as endogenous demand shocks. - Global vs. Local Determinacy:
The analysis goes beyond local determinacy (near the targeted steady state) to consider global determinacy (across the entire state space). The authors find that achieving global determinacy is more challenging, and standard policy rules often fail to ensure it. - Policy Implications:
Standard monetary policy rules, including those that respond to output-gap fluctuations or incorporate inertial behavior, cannot resolve this indeterminacy. To implement a unique equilibrium, policy must stabilize endogenous fluctuations in the natural rate of interest. - Alternative Policy Rules:
The authors suggest that to prevent non-fundamental fluctuations and ensure a unique equilibrium, monetary policy must adjust the nominal rate one-for-one with endogenous fluctuations in the natural rate. This approach is analogous to tracking the flexible-price real interest rate in RANK models to neutralize exogenous demand shocks.
The findings are particularly relevant in the context of recent high inflation and aggressive monetary policy responses. Acharya and Benhabib argue that simply raising rates in response to inflation may not be sufficient to anchor inflation expectations if the natural rate is not adequately tracked.
Global Indeterminacy in HANK Economies
Authors: Sushant Acharya, Jess Benhabib
From: Bank of Canada, New York University
The forecasting accuracy of HANK models
In this paper, Acharya et al. provide a toolkit for efficient online estimation of Heterogeneous Agent New Keynesian (HANK) models using Sequential Monte Carlo (SMC) methods. By online estimation the authors refer to using the particles describing the posterior distribution computed for the estimation up to 2000Q1 to jump-start the estimation with one or more quarters of data, thereby making it considerably faster. Here are the main findings and contributions of the paper:
- Forecasting Accuracy Comparison:
The authors compare the out-of-sample forecasting accuracy of a prominent HANK model (Bayer et al., 2022) with that of the representative agent (RANK) model (Smets and Wouters 2007, SW) . They find that the HANK model’s accuracy for real activity variables, particularly consumption growth, is notably inferior to that of the SW model. This is disappointing given that the main difference between RANK and HANK models is the replacement of the representative agent Euler equation with the aggregation of individual households’ consumption policy functions, which should theoretically better capture inequality. - Computational Challenges:
The paper highlights the computational challenges associated with estimating HANK models, particularly the large state-space that includes the distribution of wealth across households.
Traditional Markov Chain Monte Carlo (MCMC) methods are not efficient for this task due to their inability to be parallelized and the need to start from scratch for each new estimation. - Sequential Monte Carlo (SMC) Methods:
The authors propose using SMC methods, which can be parallelized and allow for “online” estimation. This means that the estimation can be updated incrementally as new data becomes available, without starting from scratch. SMC methods make it feasible to estimate models even when each likelihood computation is time-consuming. - Estimation Results:
The paper presents the results of the estimation exercise, showing that the HANK model’s forecasting performance for some series, such as inflation, is similar to that of the SW model. However, for other series, notably consumption growth, the HANK model performs worse.
The authors suggest that one key reason for the relatively worse performance of the HANK model is that many parameters affecting the model’s steady state are still calibrated rather than estimated, due to the high computational cost of recomputing the steady state.
Estimating HANK for Central Banks
Authors: Sushant Acharya, William Chen, Marco Del Negro, Keshav Dogra, Aidan Gleich, Shlok Goyal, Ethan Matlin, Donggyu Lee, Reca Sarfati, Sikata Sengupta
From: Bank of Canada, Federal Reserve Bank of New York, Harvard, MIT
