Professor Hong's research interests include model specification testing, nonlinear time series analysis, locally stationary time series analysis, generalized spectral analysis, financial econometrics, and modeling interval-valued time series data.
On specification testing, Professor Hong develops a class of sophisticated semiparametric tests for econometric models of cross-sectional, time series, and panel data respectively. The basic idea is to compare a null econometric model with a flexible nonparametric alternative, developing tests that have power against various model misspecifications. Nonparametric tools used include orthogonal series, kernel, and wavelet methods. In Hong and White (1995, Econometrica), a generalized F test is developed by comparing the sums of squared residuals of a parametric regression model and a nonparametric series regression model where the order of the series expansion grows with the sample size, ensuring the test able to detect various functional form misspecifications. This methodology is extended in Hong (1996, Econometrica) to test a dynamic regression model. This is achieved by comparing a nonparametric kernel estimator for the spectral density of the estimated model residual with the flat spectrum of a serially uncorrelated white noise. An optimal kernel function or weighting function for lags is derived to ensure that the proposed test has optimal power. This test is generalized in Hong and Kao (2004, Econometrica) to a panel data regression model where wavelets are used in nonparametric spectral density estimation. Hong and Lee (2013, Annals of Statistics) show that a loss function-based specification testing approach is asymptotically more efficient than a generalized likelihood ratio test approach which includes the tests based on comparing sums of squared residuals.
Another main area of Professor Hong's research interests is nonlinear time series analysis, locally stationary time series analysis, and generalized spectral analysis. Observing that many economic and financial time series are serially uncorrelated but not serially independent, Hong (1998, Journal of Royal Statistical Society, Series B; 2000, Journal of Rayal Statistical Society, Series B) develops nonparametric Hoeffding-type measures of and tests for serial dependence in a time series which can detect subtle dependence structure. In particular, Hong and White (2005, Econometrica) develop a challenging asymptotic distribution theory for smoothed nonparametric entropy measures of serial dependence which was not previously available in the literature. Chen and Hong (2012, Econometrica) propose a new approach to testing parameter constancy of a time series regression model against smooth structural changes as well as abrupt structural breaks.
On another important development, Hong (1999, Journal of American Statistical Association) proposes a new analytic tool for nonlinear economic time series -- the generalized spectrum. The basic idea is to transform original time series data via a complex-valued exponential function and then consider the spectrum of the transformed series. This can capture both linear and nonlinear serial dependence while avoiding the drawbacks of the conventional spectrum, which cannot capture nonlinear serial dependence, and higher order spectra (e.g., bispectrum), which requires the existence of restrictive moment conditions. Real data applications (e.g., Hong and Lee (2003a, Review of Economics and Statistics)) show that the generalized spectral tool can detect dynamic structures which would otherwise be neglected by conventional tools, thus offering new insights into economic and financial time series data. The generalized spectrum is also used to develop powerful procedures for nonlinear time series analysis. For example, Hong and Lee (2003b, Econometric Theory) use it to check any neglected dependence structure in the estimated standardized residuals of a nonlinear time series model, and Hong and Lee (2005, Review of Economic Studies) use the first order partial derivative of the generalized spectrum to focus on neglected nonlinearity in the conditional mean dynamics of a time series model. Hong and Lee (2003b, Econometric Theory) win the Koopman Econometrics Prize 2006.
Hong also works on financial econometrics. Hong and Li (2005, Review of Financial Studies) develop a nonparametric specification test for continuous-time models using discretely sampled data. The basic idea is to consider transformed data via the model-implied dynamic transition density, which should be independent and uniformly distributed if the continuous-time model is correctly specified. The proposed test is generally applicable and robust to persistent dependence in data because the i.i.d. property holds even if the original data display highly persistent dependence. Hong, Tu and Zhou (2007, Review of Financial Studies) develop copula-based tests for asymmetric dependence in asset returns and assess their economic implications.
Professor Hong has recently started a research on modeling interval-valued time series data. An interval-valued observation in a time period contains more information than a point-valued observation in the same time period. Examples of interval data include the maximum and minimum temperatures in a day, the maximum and minimum GDP growth rates in a year, the maximum and minimum stock prices in a trading day, the ask and bid prices in a trading period, the long term and short term interest, and the 90%-tile and 10%-tile incomes of a cohort in a year, etc. Interval forecasts may be of direct interest in practice, as it contains information on the range of variation and the level of economic variables. Moreover, the informational advantage of interval data can be exploited for more efficient econometric estimation and inference. Hong and his coauthors propose a new class of autoregressive conditional interval (ACI) models for interval-valued time series data. A minimum distance estimation method is developed to estimate the parameters of an ACI model. Both simulation and empirical studies show that the use of interval time series data can provide more accurate estimation for model parameters in terms of mean squared error criterion.
Contact Information
Yongmiao Hong
424 Uris Hall
Ithaca, NY 14853
yh20@cornell.edu
607-255-5130