Predictive density construction and accuracy testing with multiple possibly misspecified diffusion models☆
Introduction
Correct specification of models describing dynamics of financial assets is crucial for everything from pricing bonds and derivative assets to designing appropriate hedging strategies. Hence, it is of little surprise that there has been considerable attention given to the issue of testing for the correct specification of diffusion models. In this paper, we do not construct specification tests in the usual sense, but instead assume that all models are (possibly) misspecified and outline a simulation-based methodology for comparing the accuracy of predictive densities based on alternative models.
To place this paper in the correct historical context, note that a first generation of specification testing papers, initiated by the work of Aït-Sahalia (1996), compares the marginal densities implied by hypothesized null models with nonparametric estimates thereof, for the case of one-factor models (see also Pritsker (1998) and Jiang (1998)). While one-factor models may in some cases provide a reasonable representation for short-term interest rates, there is a somewhat widespread consensus that stock returns and term structures are better modeled using multifactor diffusions. To take this into account, Corradi and Swanson (2005a) outline a test for comparing the cumulative distribution (marginal or joint) implied by a hypothesized null model with the corresponding empirical distribution. Their test can be used in the context of multidimensional and/or multifactor models. Needless to say, tests based on the comparison of marginal distributions have no power against iid alternatives with the same marginal, while tests based on the comparison of joint distributions do not suffer from this problem. Nevertheless, correct specification of the joint distribution is not equivalent to that of the conditional; and hence focus in the literature now centers on comparing conditional distributions. When considering conditional distributions, a key difficulty that arises stems from the fact that knowledge of the drift and variance terms of a diffusion process does not in turn imply knowledge of the transition density, in general. Indeed, if the functional form of the transition density were known, one could test the hypothesis of correct specification of a diffusion via the probability integral transform approach of Diebold et al. (1998); the cross-spectrum approach of Hong (2001), Hong et al. (2002) and Hong and Li (2005); the martingalization-type Kolmogorov test of Bai (2003); or via the normality transformation approaches of Bontemps and Meddahi (2005) and Duan (2003). Furthermore, for the case in which the transition density is unknown, tests could be constructed by comparing the kernel (conditional) density estimator of the actual and simulated data, as in Altissimo and Mele (2009) and Thompson (2008); by comparing the conditional distribution of the simulated and of the historical data, as in Bhardwaj et al. (2008); or by using the approaches of Aït-Sahalia (2002) and Aït-Sahalia et al. (2009), where closed form approximations of conditional densities under the null are compared with data-driven kernel density estimates.
All of the papers cited above deal with testing for the correct specification of a given diffusion model. Nevertheless, and as alluded to above, we believe that all models are probably best viewed as approximations of reality and, thus, are likely to be misspecified. Therefore, we focus on choosing the “best” model from amongst (multiple) misspecified alternatives. Moreover, the “best” model is selected by constructing tests that compare both predictive densities and/or predictive conditional confidence intervals associated with alternative models.
Our approach is to measure accuracy using a distributional generalization of mean square error, as defined in Corradi and Swanson (2005b). Namely, let be the distribution of given , evaluated at , implied by diffusion model , and let be the distribution associated with the underlying and unknown “true” model. Now, choose model over model 1, say, if . Our tests can be viewed as distributional generalizations of both Diebold and Mariano (1995) and White (2000). Note that if we knew in closed form, then we could proceed as in Corradi and Swanson, 2006a, Corradi and Swanson, 2006b. However, the functional form of the model implied conditional distribution is unknown in closed form, in general, and hence we rely on a simulation-based approach to facilitate testing. As is customary in the out-of-sample evaluation literature, the sample of observations is split into two subsamples, such that , where only the last observations are used for predictive evaluation. We first simulate -step ahead paths, using as starting values. Then, a scaled difference between the conditional distribution, estimated with historical as well as simulated data, is used to construct our test statistic. One complication that arises in this setup is that for the case of stochastic volatility (SV) models, the initial value of the volatility process is unobserved. To overcome this problem, it suffices to simulate the process using different random initial values for the volatility process. Thereafter, one simply constructs the empirical distribution of the asset price process for any given initial value of the volatility process and takes an average over the latter. This integrates out the effect of the volatility initial value.
The limiting distributions of the suggested statistics are shown to be (functionals of) Gaussian processes with covariance kernels that reflect the contribution of recursive parameter estimation error. In order to provide asymptotically (first-order) valid critical values, we introduce a new bootstrap procedure that mimics the contribution of parameter estimation error in a recursive setting. This is achieved by establishing consistency and asymptotic normality of nonparametric simulated quasi maximum likelihood (NPSQML) estimators of (possibly misspecified) diffusion models, in a recursive setting, and by establishing the first-order validity of their bootstrap analogs.
Of final note is that we test the same null hypothesis as Corradi and Swanson (2006a), and we estimate empirical conditional distributions using both historical and simulated data, as in Bhardwaj et al. (2008). However, there are many differences between those papers and this one. Five such differences are the following. First, we show the asymptotic equivalence of recursive NPSQMLE (Nonparametric Simulated Quasi Maximum Likelihood Estimators) and recursive QMLE. Second, we show the asymptotic equivalence of recursively estimated NPSQMLE and recursive QMLE for partially unobservable multidimensional diffusions (e.g. for stochastic volatility models). This extends in a non-trivial manner the NPSQMLE of Fermanian and Salanié (2004). Third, we establish the first order validity of bootstrap critical values for recursive NPSQMLE, in the case of both observable and partially unobservable diffusions. To the best of our knowledge, there are no available results on bootstrapping NPSQMLE. Fourth, we allow for jumps in the return process, and we recursively estimate the intensity and the parameters of the jump size density. Finally, we develop Diebold–Mariano type Reality Check tests for cases where (a) the CDF is not known in closed form, and (b) data are generated by partially unobservable jump diffusion processes.
The rest of the paper is organized as follows. In Section 2, we define the setup. Section 3 outlines the testing procedure for choosing between models and establishes the asymptotic properties thereof. In Section 4, we develop a recursive version of the NPSQML estimator of Fermanian and Salanié (2004) and outline conditions under which asymptotic equivalence between NPSQML and the corresponding recursive QMLE obtains. An empirical illustration is provided in Section 5, in which various models of the effective federal funds rate are compared. All proofs are collected in an Appendix. Hereafter, let denote the probability law governing the resampled series, conditional on the (entire) sample, let and denote the mean and variance operators associated with . Further, let denote a term converging to zero in -probability, conditional on the sample except a subset of probability measure approaching zero. Finally, let denote a term which is bounded in -probability, conditional on the sample, and for all samples except a subset with probability measure approaching zero.
Section snippets
Set-up
First, consider one factor jump diffusion models. Namely, for consider1: where is a Poisson process with intensity parameter , finite, and the jump size, , is iid with marginal distribution given by . Both and are assumed to be independent of the driving
One factor models
First, consider comparing the predictive accuracy of two possibly misspecified diffusion models. The hypotheses of interest are: Notice that the hypotheses are stated using a particular interval (i.e., ) so that the objective is evaluation of predictive densities for a given range of values. The test statistic is:
Recursive nonparametric simulated quasi maximum likelihood estimators
In this section we develop a recursive version of the nonparametric simulated (quasi) maximum likelihood (NPSQML) estimator of Fermanian and Salanié (2004) and outline conditions under which asymptotic equivalence between the NPSQML estimator and the corresponding recursive QML estimator obtains, hence ensuring that A4 and A4′ hold. Analogous results are also established for the bootstrap counterpart of the recursive NPSQML estimators.
A previous version of this paper contains results analogous
Empirical illustration: choosing between CIR, SV, and SVJ models
In this section, we choose between Cox–Ingersoll–Ross (CIR), stochastic volatility (SV) and stochastic volatility with jumps (SVJ) models by comparing the models’ predictive performance across two different sample periods. Our primary objective is to illustrate the implementation of our test statistics and our secondary objective is to assess whether the choice of model is impacted by the choice of sample period. There are many precedents in the empirical literature suggesting that evaluation
References (40)
- et al.
Testing normality: a GMM approach
Journal of Econometrics
(2005) - et al.
Spectral GMM estimation of continuous-time Processes
Journal of Econometrics
(2003) - et al.
Bootstrap tests for diffusion processes
Journal of Econometrics
(2005) - et al.
Predictive density and conditional confidence interval accuracy tests
Journal of Econometrics
(2006) - et al.
Predictive density evaluation
Ergodicity and exponential -mixing bounds for multivariate diffusions with jumps
Stochastic Processes and Their Applications
(2007)Estimation of affine asset pricing models using empirical characteristic function
Journal of Econometrics
(2001)- et al.
Simulated nonparametric estimation of dynamic models with application in finance
Review of Economic Studies
(2009) Testing continuous time models of the spot interest rate
Review of Financial Studies
(1996)Maximum likelihood estimation of discretely sampled diffusions: a closed form approximation approach
Econometrica
(2002)
Nonparametric transition-based tests for diffusions
Journal of the American Statistical Association
Testing parametric conditional distributions of dynamic models
Review of Economics and Statistics
A simulation based specification test for diffusion processes
Journal of Business and Economic Statistics
Efficient estimation of general dynamic models with a continuum of moment conditions
Journal of Econometrics
A test for comparing multiple misspecified conditional intervals
Econometric Theory
Nonparametric bootstrap procedures for predictive inference based on recursive estimation schemes
International Economic Review
Evaluating density forecasts with applications to finance and management
International Economic Review
Comparing predictive accuracy
Journal of Business and Economic Statistics
Mixing Properties and Examples
Cited by (8)
Testing for jumps and jump intensity path dependence
2018, Journal of EconometricsCitation Excerpt :Misspecification of one or both components will invalidate the test. Additionally, the likelihood function of a jump diffusion is not generally known in closed form, and therefore estimation (which is needed for test statistic construction) is usually based on either simulated GMM (see Duffie and Singleton (1993) and Andersen et al. (2002)), indirect inference (see Gourieroux et al. (1993) and Gallant and Tauchen (1996)), or nonparametric simulated maximum likelihood (see Fermanian and Salanié (2004) and Corradi and Swanson (2011)). However, it goes without saying that one cannot simulate a diffusion with a negative intensity parameter.
Consistent nonparametric specification tests for stochastic volatility models based on the return distribution
2017, Journal of Empirical FinanceCitation Excerpt :However, the unobservability of the volatility process in (1) makes the aforementioned research not directly applicable. Corradi and Swanson (2011), henceforth CS11, consider a conditional distribution based nonparametric test for stochastic volatility models, where the authors assume the observed series to be strictly stationary. In our model we assume the observed return series (first difference of the observed series) to be stationary and we allow the observed series to exhibit unit-root type dynamics.
Nonparametric specification tests for stochastic volatility models based on volatility density
2015, Journal of EconometricsCitation Excerpt :Early contributions include the moment restrictions based tests (such as Gallant et al., 1997). Recent contributions include Corradi and Distaso (2006), who propose testing the specification of stochastic volatility models based on the moment information of realized volatility measures, and Corradi and Swanson (2011), who propose basing the test on the one-step predictive density of observed series. Zu and Boswijk (2009) propose tests based on the density function and distribution function of the observed returns.
Prediction and simulation using simple models characterized by nonstationarity and seasonality
2015, International Review of Economics and FinanceCitation Excerpt :The starting values for the simulations were fixed to be the last observation of the historical sample used in estimation. In this sense, following the theoretical arguments made in Corradi and Swanson (2011), simulated distributions for a given future period, say h, are asymptotically equivalent to h-step ahead predictive densities. Note that the models used for simulation are the models outlined above (see Section 2.1), including an error term, where the error is assumed to be iid normal, and where the variance of the error term is estimated using the residuals of the fitted model.
In- and out-of-sample specification analysis of spot rate models: Further evidence for the period 1982-2008
2011, Journal of Empirical FinanceCitation Excerpt :Concluding remarks are given in Section 6. Our testing methodology follows that of Bhardwaj et al. (2008) and Corradi and Swanson (2011). In particular, we first implement in-sample consistent specification tests that due to Bhardwaj et al. (2008).
A test for multidimensional reducible diffusion models
2020, Vibroengineering Procedia
- ☆
Corradi gratefully acknowledges ESRC grant RES-000-23-0006 and RES-062-23-0311, and Swanson acknowledges financial support from a Rutgers University Research Council grant. We would like to thank the co-editor, Ron Gallant, a referee, Federico Bandi, Marine Carrasco, Javier Hidalgo, Antonio Mele, Andrew Patton, Eric Renault, John Rust and the seminar participants at the annual joint CORE, ECARES, and KU Leuven econometrics workshop, the 2007 Summer Meeting of the Econometric Society, the Marseille conference in honor of Russell Davidson, as well as faculty at the following universities: University of Montreal, LSE, Michigan State University, New York University, University of Chicago GBS, and the University of Maryland, for their useful comments on earlier versions of this paper. Additionally, we would like to thank Lili Cai for excellent research assistance.