发布时间：2018-08-13 | 来源：数学院南楼N204
An Averaging GMM Estimator Robust to Model Misspecification
【Abstract】This paper studies the averaging generalized method of moments (GMM) estimator that combines a conservative GMM estimator based on valid moment conditions and an aggressive GMM estimator based on both valid and possibly misspecified moment conditions, where the weight is the sample analog of an infeasible optimal weight. We establish asymptotic theory on uniform approximation of the upper and lower bounds of the finite-sample truncated risk difference between two estimators, which is used to compare the averaging GMM estimator and the conservative GMM estimator. Under some sufficient conditions, we show that the asymptotic lower bound of the truncated risk difference between the averaging estimator and the conservative estimator is strictly less than zero, while the asymptotic upper bound is zero uniformly over any degree of misspecification. Extending seminal results on the James-Stein estimator, this uniform dominance is established in non-Gaussian semiparametric nonlinear models. The simulation results support our theoretical findings.
Solving Asset Pricing Models via 2 Stage Penalized B-spline Regression
【Abstract】This study proposes a novel nonparametric estimation approach to solving structural asset pricing models, which allows the true dynamics of state variables to determine equilibrium asset prices. Unlike most numerical solution methods, our method offers a more robust estimate of the model solution for asset prices without misspecification errors about the true underlying processes of state variables or the form of unknown functions while also taking into account investors’ preferences. Through a 2 stage penalized B-spline regression, we establish the asymptotics of the estimation for a broad class of stationary Markov state variables. Our estimator overcomes the ill-posed inverse problem which is typical in nonparametric instrumental variable regression, and achieves the optimal convergence rate. We design a fast generalized cross-validation procedure to tune the penalty parameter effectively for ease of practical use. In addition to being robust to the choice of the spline basis, our approach exhibits superior accuracy in small samples. As an application, we estimate a misspecification-free implied dividend yield from a rational model and re-investigate return predictability. We find that high implied dividend yield significantly predicts lower future cash flows and higher interest rates at short horizons; however, we find no evidence for its ability to forecast returns in the period from 1947-2017.
Transfer Optimization & Multi-Task Optimization
What Do A Billion Observations Say About Distance and Relationship Lending?
【Abstract】Using one billion observations on the locations of bank branches and firms in China, we find strong evidence of a novel U-shaped relationship between lender-borrower distance and soft information. Lending intensities decrease with distance within a short range but increase with distance beyond that. Distant borrowers have fewer explicit third-party loan guarantees but provide more implicit guarantees from their connected firms that borrow from the same bank. This firm network facilitates lenders to obtain soft information and manage risks. The default ratios are significantly lower for firms that are either nearby or distant from the lending bank branches. Moreover, we observe directly the lenders’ soft information by tracing out whether banks downgrade internal loan ratings before the delinquency. Banks can predict the delinquency more accurately for the firms in either short or long distance.
An Unobserved Component Modeling Approach to Evaluating Multi-horizon Forecasts
【Abstract】We propose a state space modeling framework to evaluate a set of forecasts that target the same variable but are updated along the forecast horizon. The approach decomposes forecast errors into three distinct horizon-specific processes, namely, bias, rational error and implicit error, and attributes forecast revisions to corrections for these forecast errors. We conduct Monte Carlo simulations to show the ability of the modeling approach to identify the correct error compositions across horizons. By evaluating multi-horizon daily maximum temperature forecasts for Melbourne, Australia, we demonstrate how this modeling framework analyzes the dynamics of the forecast revision structure across horizons. Understanding forecast revisions is critical for weather forecast users to determine the optimal timing for their planning decision.
Forecasting Stock Returns: Some New Evidence
【Abstract】We develop a novel method to impose constraints on univariate predictive regressions of stock returns. Unlike the previous approaches in the literature, we implement our constraints directly on the predictor, setting it to zero whenever its value falls within the variable’s past 24-month high and low. Empirically, we find that relative to standard unconstrained predictive regressions, our approach leads to significantly larger forecasting gains. We also show how a simple equal-weighted combination of our constrained forecasts leads to further improvements in forecast accuracy, generating forecasts that are more accurate than those obtained using existing constrained methods. Further analysis confirms that these findings are robust to the presence of model instabilities and structural breaks.