Macro-Financial Aggregates in a DSGE Framework (2017)

Paper (PDF)

This paper studies the main aggregates of the financial sector in the United States and their relationship with macroeconomic aggregates in the context of a medium-scale dynamic stochastic general equilibrium model. Aggregate leverage in the financial industry, corporate credit spread and the aggregate balance sheet composition are the financial industry aggregates added to the general equilibrium framework. The model features a financial accelerator derived from a moral hazard problem between the financial industry and households. In order to account for risk-aversion effects, the solution method involves a second order approximation that induces a non-linear state-space representation. A self-tuning marginalized particle filter algorithm is proposed on the pruned state-space system to obtain parameter estimates. Using the hybrid approach to estimating a DSGE model from the raw data described in Canova (2014), I decompose the main series of the financial sector and the general economy into two orthogonal components, one DSGE-model explained and a non-model related process.  As an application, I perform a crisis impact exercise by analyzing the generalized impulse response function under different levels of aggregate bank capitalization and leverage. These responses are non-symmetric, non-scalable and state-dependent, in line with the findings by Andreasen et al. (2016).


Fed A.I. (2017)

(with George Lentzas and Rohit Suvarna)


We study how a Q-learning algorithm is able to learn the benevolent planner monetary policy from a dynamic stochastic general equilibrium model (DSGE). We simulate paths of the economy and make the agent "play" the economy to learn the best path of interest rates. We show that in the context of a medium scale DSGE model, the algorithm successfully increments the average reward per epochs and generates actions that like the Taylor rule respond to inflation and output gap, but also to additional variables in the economy.



Data-dependent Reproducing Kernel Hilbert Space Splines for Low Frequency Volatility (2017)

(with Timothy Copeland)

This paper presents a framework to estimate the low-frequency component of the volatility of returns in the context of multiplicative GARCH volatility models.  We show that the spline-GARCH model can be seen a particular family of a wide class of estimators defined by a Reproducing Kernel Hilbert Space with different metrics that act as penalties over the curve shape. Data-dependent kernels have an advantage to the claim that they capture low-frequency components related to economy fundamentals, unlike typical smoothing splines.


Factor Copula for Dynamic Asset Allocation (2017)

(with Susy Pan and Ruonan Yang)

We develop an estimation framework for robust dynamic asset allocation that internalizes the macroeconomic dependency of asset class distributions by constructing a factor copula model.


Kernel Pricing with multi-class assets (2017)

This work estimates a kernel pricing general equilibrium model using data from stocks, bond market, foreign exchange and derivatives using MCMC methods over a non-linear solution of the model.


Mortgage Hedging in Interest Rate Swaps: a General Equilibrium Model Estimation (2017)

(with Jun Peng)

This work quantifies the mortgage hedging component in interest rate swaps using a macroeconomic general equilibrium model with a financial sector. In an economy where financial intermediaries hold vast amounts of mortgage-backed securities, prepayment risk entails an interest rate exposure that banks might hedge in IRS markets.