Search

We propose a procedure for estimating the critical values of the extended Kolmogorov- Smirnov tests of First and Second Order Stochastic Dominance in the general K-prospect case. We allow for the observations to be serially dependent and, for the first time, we can accommodate general dependence amongst the prospects which are to be ranked. Also, the prospects may be the residuals from certain conditional models, opening the way for conditional ranking. We also propose a test of Prospect Stochastic Dominance. Our method is subsampling; we show that the resulting tests are consistent and powerful against some N|1/2 local alternatives even when computed with a data-based subsample size. We also propose some heuristic methods for selecting subsample size and demonstrate in simulations that they perform reasonably. We show that our test is asymptotically similar on the entire boundary of the null hypothesis, and is unbiased. In comparison, any method based on resampling or simulating from the least favorable distribution does not have these properties and consequently will have less power against some alternatives.

We investigate a new separable nonparametric model for time series, which includes many autoregressive conditional heteroskedastic (ARCH) models and autoregressive (AR) models already discussed in the literature. We also propose a new estimation procedure called LIVE, or local instrumental variable estimation, that is based on a localization of the classical instrumental variable method. Our method has considerable computational advantages over the competing marginal integration or projection method. We also consider a more efficient two-step likelihood-based procedure and show that this yields both asymptotic and finite-sample performance gains. This paper is based on Chapter 2 of the first author's Ph.D. dissertation from Yale University. We thank Wolfgang Hardle, Joel Horowitz, Peter Phillips, and Dag Tjostheim for helpful discussions. We are also grateful to Donald Andrews and two anonymous referees for valuable comments. The second author thanks the National Science Foundation and the ESRC for financial support.

We investigate a class of estimators for Linear Regression models where the dependent variable is subject to bid-ask censoring. Our estimation method is based on a definition of error that is zero when the predictor lies between the actual bid price and ask price, and linear outside this range. Our estimator minimizes a sum of such squared errors; it is non-linear, and indeed the criterion function itself is non smooth. We establish its asymptotic properties using the approach of Pakes & Pollard (1989). We compare the estimator with mid-point OLS.

We investigate a new separable nonparametric model for time series, which includes many ARCH models and AR models already discussed in the literature. We also propose a new estimation procedure called LIVE, or local instrumental variable estimation, that is based on a localization of the classical instrumental variable method. Our method has considerable computational advantages over the competing marginal integration or projection method. We also consider a more efficient two-step likelihood-based procedure, and show that this yields both asymptotic and finite sample performance gains.

We introduce a new method for the estimation of discount functions, yield curves and forward curves from government issued coupon bonds. Our approach is nonparametric and does not assume a particular functional form for the discount function although we do show how to impose various restrictions in the estimation. Our method is based on kernel smoothing and is defined as the minimum of some localized population moment condition. The solution to the sample problem is not explicit and our estimation procedure is iterative, rather like the backfitting method of estimating additive nonparametric models. We establish the asymptotic normality of our methods using the asymptotic representation of our estimator as an infinite series with declining coefficients. The rate of convergence is standard for one dimensional nonparametric regression. We investigate the finite sample performance of our method, in comparison with other well-established methods, in a small simulation experiment.

This paper estimates the implied stochastic process of the volatility of the Swiss market index (SMI) from the prices of options written on it. A GARCH(1,1) model is shown to be a good parameterization of the process. Then, using the GARCH option pricing model of Duan (1991), the implied volatility process is estimated by a simulation minimization method from option price data. We find the persistence of volatility shocks implied by options on the SMI to be very close to that estimated from historical data on the index itself. Comparing the performances of the implied GARCH option pricing model to that of the Black and Scholes model it appears that the overall pricing performance of the former is superior. However the large sample standard deviations of the out-of-sample pricing errors suggest that this result should be taken with caution.

We show that the recently developed nonparametric procedure for fitting the term structure of interest rates developed by Linton, Mammen, Nielson and Tanggaard (2000) overall performs notably better than the highly flexible McCulloch (1975) cubic spline and Fama and Bliss (1987) bootstrap methods. However, if interest is limited to the Treasury bill region alone then the Fama-Bliss method demonstrates superior performance. We further show, via simulation, that using the estimated short rate from Linton-Mammen-Nielson-Tanggaard procedure as a proxy for the short rate has higher precision than the commonly used proxies of the one and three month Treasury bill rates. It is demonstrated that this precision is important when using proxies to estimate the stochastic process governing the evolution of the short rate.

Estimation of heteroskedasticity and autocorrelation consistent covariance matrices (HACs) is a well established problem in time series. Results have been established under a variety of weak conditions on temporal dependence and heterogeneity that allow one to conduct inference on a variety of statistics, see Newey and West (1987), Hansen (1992), de Jong and Davidson (2000), and Robinson (2004). Indeed there is an extensive literature on automating these procedures starting with Andrews (1991). Alternative methods for conducting inference include the bootstrap for which there is also now a very active research program in time series especially, see Lahiri (2003) for an overview. One convenient method for time series is the subsampling approach of Politis, Romano, andWolf (1999). This method was used by Linton, Maasoumi, andWhang (2003) (henceforth LMW) in the context of testing for stochastic dominance. This paper is concerned with the practical problem of conducting inference in a vector time series setting when the data is unbalanced or incomplete. In this case, one can work only with the common sample, to which a standard HAC/bootstrap theory applies, but at the expense of throwing away data and perhaps losing effciency. An alternative is to use some sort of imputation method, but this requires additional modelling assumptions, which we would rather avoid.1 We show how the sampling theory changes and how to modify the resampling algorithms to accommodate the problem of missing data. We also discuss effciency and power. Unbalanced data of the type we consider are quite common in financial panel data, see for example Connor and Korajczyk (1993). These data also occur in cross-country studies.

We investigate the model of Froot and Stein (1998), a model that has very strong implications for risk management. We argue that their conclusions are too strong and need to be qualified. Also, there are some unusual consequences of their model, which may be linked to the chosen pricing formula.

We consider nonparametric regression in a marginal longitudinal data framework. Previous work ([3]) has shown that the kernel nonparametric regression methods extant in the literature for such correlated data have the discouraging property that they generally do not improve upon methods that ignore the correlation structure entirely. The latter methods are called working independence methods. We construct a two- stage kernel-based estimator that asymptotically uniformly improves upon the working independence estimator. A small simulation study is given in support of the asymptotics.

We develop inference tools in a semiparametric partially linear regression model with missing response data. A class of estimators is defined that includes as special cases: a semiparametric regression imputation estimator, a marginal average estimator and a (marginal) propensity score weighted estimator. We show that any of our class of estimators is asymptotically normal. The three special estimators have the same asymptotic variance. They achieve the semiparametric efficiency bound in the homoskedastic Gaussian case. We show that the Jackknife method can be used to consistently estimate the asymptotic variance. Our model and estimators are defined with a view to avoid the curse of dimensionality, that severely limits the applicability of existing methods. The empirical likelihood method is developed. It is shown that when missing responses are imputed using the semiparametric regression method the empirical log-likelihood is asymptotically a scaled chi-square variable. An adjusted empirical log-likelihood ratio, which is asymptotically standard chi-square, is obtained. Also, a bootstrap empirical log-likelihood ratio is derived and its distribution is used to approximate that of the imputed empirical log-likelihood ratio. A simulation study is conducted to compare the adjusted and bootstrap empirical likelihood with the normal approximation based method in terms of coverage accuracies and average lengths of confidence intervals. Based on biases and standard errors, a comparison is also made by simulation between the proposed estimators and the related estimators.

We apply semiparametric efficient estimation procedures for a seemingly unrelated regression model where the multivariate error density is elliptically symmetric to study the efficiency of the foreign exchange market. We consider both cointegrating regressions and standard stationary regressions. The elliptical symmetry assumption allows us to avoid the curse of dimensionality problem that typically arises in multivariate semiparametric estimation procedures, because the multivariate elliptically symmetric density function can be written as a function of a scalar transformation of the observed multivariate data. We test the unbiasedness hypothesis on both weekly and daily exchange rate data and strongly reject unbiasedness at the weekly horizon, but fail to reject the unbiasedness hypothesis on the daily data. Estimates of the semiparametric procedure in some cases differ substantially from traditional OLS estimates.

In this note we propose a simple method of measuring directional predictability and testing for the hypothesis that a given time series has no directional predictability. The test is based on the correlogram of quantile hits. We provide the distribution theory needed to conduct inference, propose some model free upper bound critical values, and apply our methods to stock index return data. The empirical results suggests some directional predictability in returns especially in mid range quantiles like 5%-10%.

For vectors z and w and scalar v, let r(v, z, w) be a function that can be non-parametrically estimated consistently and asymptotically normally, such as a distribution, density, or conditional mean regression function. We provide consistent, asymptotically normal nonparametric estimators for the functions G and H, where r(v, z, w) = H[vG(z), w], and some related models. This framework encompasses homothetic and homothetically separable functions, and transformed partly additive models r(v, z, w) = h[v + g(z), w] for unknown functions g and h. Such models reduce the curse of dimensionality, provide a natural generalization of linear index models, and are widely used in utility, production, and cost function applications. We also provide an estimator of G that is oracle efficient, achieving the same performance as an estimator based on local least squares when H is known.

For vectors x and w, let r(x,w) be a function that can be nonparametrically estimated consistently and asymptotically normally. We provide consistent, asymptotically normal estimators for the functions g and h, where r(x,w) = h[g(x), w], g is linearly homogeneous and h is monotonic in g. This framework encompasses homothetic and homothetically separable functions. Such models reduce the curse of dimensionality, provide a natural generalization of linear index models, and are widely used in utility, production, and cost function applications. Extensions to related functional forms include a generalized partly linear model with unknown link function. We provide simulation evidence on the small sample performance of our estimator, and we apply our method to a Chinese production dataset.

We investigate a class of semiparametric ARCH(∞) models that includes as a special case the partially nonparametric (PNP) model introduced by Engle and Ng (1993) and which allows for both flexible dynamics and flexible function form with regard to the 'news impact' function. We propose an estimation method that is based on kernel smoothing and profiled likelihood. We establish the distribution theory of the parametric components and the pointwise distribution of the nonparametric component of the model. We also discuss efficiency of both the parametric and nonparametric part. We investigate the performance of our procedures on simulated data and on a sample of S&P500 daily returns. We find some evidence of asymmetric news impact functions in the data.

For vectors x and w, let r(x,w) be a function that can be nonparametrically estimated consistently and asymptotically normally. We provide consistent, asymptotically normal estimators for the functions g and h, where r(x,w)=h[g(x),w], g is linearly homogeneous and h is monotonic in g. This framework encompasses homothetic and homothetically separable functions. Such models reduce the curse of dimensionality, provide a natural generalization of linear index models, and are widely used in utility, production, and cost function applications. Extensions to related functional forms include a generalized partly linear model with unknown link function and endogenous regressors. We provide simulation evidence on the small sample performance of our estimator, and we apply our method to a Chinese production dataset.

We examine the relationship between the risk premium on the Center for Research on Security Prices (CRSP) value-weighted index total return and its conditional variance. We propose a new semiparametric model in which the conditional variance process is parametric and the conditional mean is an arbitrary function of the conditional variance. For monthly CRSP value-weighted excess returns, the relationship between the two moments that we uncover is nonlinear and nonmonotonic.

We investigate a class of semiparametric ARCH(infinity) models that includes as a special case the partially nonparametric (PNP) model introduced by Engle and Ng (1993) and which allows for both flexible dynamics and flexible function form with regard to the 'news impact' function. We show that the functional part of the model satisfies a type II linear integral equation and give simple conditions under which there is a unique solution. We propose an estimation method that is based on kernel smoothing and profiled likelihood. We establish the distribution theory of the parametric components and the pointwise distribution of the nonparametric component of the model. We also discuss efficiency of both the parametric and nonparametric part. We investigate the performance of our procedures on simulated data and on a sample of S&P500 index returns. We find some evidence of asymmetric news impact functions in the daily and weekly data, but some significant differences from the usual shape found in purely parametric analysis.

We investigate the model of Froot & Stein (1998), a model which has very strong implications for risk management. We argue that their conclusions are too strong and need to be qualified. We also argue that their analysis is incorrect and incomplete. Specifically, there are some unusual consequences of their model, which may be linked to the chosen pricing formula.

We study a very general setting, and propose a procedure for estimating the critical values of the extended Kolmogorov-Smirnov tests of First and Second Order Stochastic Dominance due to McFadden (1989) in the general k-prospect case. We allow for the observations to be generally serially dependent and, for the first time, we can accommodate general dependence amongst the prospects which are to be ranked. Also, the prospects may be the residuals from certain conditional models, opening the way for conditional ranking. We also propose a test of Prospect Stochastic Dominance. Our method is based on subsampling and we show that the resulting data tests are consistent.

We provide easy to verify sufficient conditions for the consistency and asymptotic normality of a class of semiparametric optimization estimators where the criterion function does not obey standard smoothness conditions and simultaneously depends on some nonparametric estimators that can themselves depend on the parameters to be estimated. Our results extend existing theories like those of Pakes and Pollard (1989), Andrews (1994a) and Newey (1994). We also show that bootstrap provides asymptotically correct confidence regions for the finite dimensional parameters. We apply our results to two examples: a 'hit rate' and a partially linear median regression with some endogenous regressors.

We introduce a kernel-based estimator of the density function and regression function for data that have been grouped into family totals. We allow for a common intrafamily component but require that observations from different families be independent. We establish consistency and asymptotic normality for our procedures. As usual, the rates of convergence can be very slow depending on the behavior of the characteristic function at infinity. We investigate the practical performance of our method in a simple Monte Carlo experiment.

This paper derives the asymptotic distribution of nonparametric neural network estimator of the Lyapunov exponent in a noisy system proposed by Nychka et al (1992) and others. Positivity of the Lyapunov exponent is an operational definition of chaos. We introduce a statistical framework for testing the chaotic hypothesis based on the estimated Lyapunov exponents and a consistent variance estimator. A simulation study to evaluate small sample performance is reported. We also apply our procedures to daily stock return datasets. In most cases we strongly reject the hypothesis of chaos; one mild exception is in some higher power transformed absolute returns, where we still find evidence against the hypothesis but it is somewhat weaker.

We propose a procedure for estimating the critical values of the extended Kolmogorov- Smirnov tests of Stochastic Dominance of arbitrary order in the general K-prospect case. We allow for the observations to be serially dependent and, for the first time, we can accommodate general dependence amongst the prospects which are to be ranked. Also, the prospects may be the residuals from certain conditional models, opening the way for conditional ranking. We also propose a test of Prospect Stochastic Dominance. Our method is based on subsampling and we show that the resulting tests are consistent and powerful against some N −1/2 local alternatives. We also propose some heuristic methods for selecting subsample size and demonstrate in simulations that they perform reasonably. We describe an alternative method for obtaining critical values based on recentering the test statistic and using full sample bootstrap methods. We compare the two methods in theory and in practice.

We propose estimators of previous termfeatures of the distributionnext term of an unobserved random variable W. What is observed is previous termanext term sample of Y,V,X where previous termanext term binary Y equals one when W exceeds previous termanext term threshold V determined by experimental design, and X are covariates. Potential applications include bioassay and destructive duration analysis. Our empirical application is referendum contingent valuation in resource economics, where one is interested in previous termfeatures of the distributionnext term of values W (willingness to pay) placed by consumers on previous termanext term public good such as endangered species. Sample consumers with characteristics X are asked whether they favor (with Y=1 if yes and zero otherwise) previous termanext term referendum that would provide the good at previous termanext term cost V specified by experimental design. This paper provides estimators for quantiles and conditional on X moments of W under both nonparametric and semiparametric specifications.

We investigate the performance of a class of semiparametric estimators of the treatment effect via asymptotic expansions. We derive approximations to the first two moments of the estimator that are valid to 'second order'. We use these approximations to define a method of bandwidth selection. We also propose a degrees- of-freedom like bias correction that improves the second order properties of the estimator but without requiring estimation of higher order derivatives of the unknown propensity score. We provide some numerical calibrations of the results.

We develop a nonparametric estimator for the volatility structure of the zero coupon yield curve in the Heath-Jarrow-Morton framework. The estimator incorporates cross-sectional restrictions along the maturity dimension, and also allows for measurement errors, which arise from the estimation of the yield curve from noisy data. The estimates are implemented with daily CRSP bond data.

We show that the recently developed non-parametric procedure for fitting the term structure of interest rates developed by Linton, Mammen, Nielsen, and Tanggaard (J Econ 105(1):185–223, 2001) overall performs notably better than the highly flexible McCulloch (J Finon 30:811–830, 1975) cubic spline and Fama and Bliss (Am Econ Rev 77:680–692, 1987) bootstrap methods. However, if interest is limited to the Treasury-bill region alone then the Fama–Bliss method demonstrates superior performance. We further show, via simulation, that using the estimated short rate from the Linton–Mammen–Nielsen–Tanggaard procedure as a proxy for the short rate has higher precision then the commonly used proxies of the one and three month Treasury-bill rates. It is demonstrated that this precision is important when using proxies to estimate the stochastic process governing the evolution of the short rate

We propose a new estimator for nonparametric regression based on local likelihood estimation using an estimated error score function obtained from the residuals of a preliminary nonparametric regression. We show that our estimator is asymptotically equivalent to the infeasible local maximum likelihood estimator [Staniswalis (1989)], and hence improves on standard kernel estimators when the error distribution is not normal. We investigate the finite sample performance of our procedure on simulated data.

We develop new tests of the capital asset pricing model (CAPM) that take account of and are valid under the assumption that the distribution generating returns is elliptically symmetric; this assumption is necessary and sufficient for the validity of the CAPM. Our test is based on semiparametric efficient estimation procedures for a seemingly unrelated regression model where the multivariate error density is elliptically symmetric, but otherwise unrestricted. The elliptical symmetry assumption allows us to avert the curse of dimensionality problem that typically arises in multivariate semiparametric estimation procedures, because the multivariate elliptically symmetric density function can be written as a function of a scalar transformation of the observed multivariate data. The elliptically symmetric family includes a number of thick-tailed distributions and so is potentially relevant in financial applications. Our estimated betas are lower than the OLS estimates, and our parameter estimates are much less consistent with the CAPM restrictions than the corresponding OLS estimates. Nous développons de nouveaux tests du modèle d'évaluation des actifs financiers (" CAPM ") qui tiennent compte de, et sont valides sous, l'hypothèse que les retours des actifs découlent d'un loi de probabilité elliptiquement symétrique. Cette hypothèse est nécessaire et suffisante pour la validité du CAPM. Notre test utilise un estimateur des paramètres du modèle qui a l'efficacité semiparamétrique quand on a un modèle de régression apparemment sans relation et qui a des erreurs qui suivent une loi elliptiquement symétrique. L'hypothèse de la symétrie elliptique nous permet d'éviter le problème d'estimer non-paramétriquement une fonction de haute dimension parce qu'on peut écrire la densité d'une loi elliptique comme une fonction d'une transformation unidimensionnelle de la variable aléatoire multidimensionnelle. La famille des lois elliptiquement symétriques inclue plusieurs lois leptokurtiques, donc elle est pertinente à des applications financières. Les bêtas obtenus avec notre estimateur sont plus bas que ceux qui sont obtenus en utilisant des moindres carrés, et sont moins compatibles avec le CAPM.

We discuss a number of issues in the smoothed nonparametric estimation of kernel conditional probability density functions for stationary processes. The kernel conditional density estimate is a ratio of joint and marginal density estimates. We point out the different implications of leading choices of bandwidths in numerator and denominator for the ability of the estimate to integrate to one and to have finite moments. Again bearing in mind different bandwidth possibilities, we discuss asymptotic theory for the estimate: asymptotic bias and variance are calculated under various conditions, an extended discussion of bandwidth choice is included, and a central limit theorem is given.

We propose new procedures for estimating the component functions in both additive and multiplicative nonparametric marker-dependent hazard models. We work with a full counting process framework that allows for left truncation and right censoring and time-varying covariates. Our procedures are based on kernel hazard estimation as developed by Nielsen and Linton and on the idea of marginal integration. We provide a central limit theorem for the marginal integration estimator. We then define estimators based on finite-step backfitting in both additive and multiplicative cases and prove that these estimators are asymptotically normal and have smaller variance than the marginal integration method.

A positive Lyapunov exponent is one pratical definition of chaos. We develop a formal test for chaos in a noisy system based on the sign of the Lyapunov exponent. The test utilizes nonparametric regression techniques including local quadratic regression and neural networks. When our procedures are applied to international real output series, the Lyapunov exponent estimates are negative and the positivity hypothesis of the expononent is significantly rejected in many cases. One possible interpretation of this result is that the traditional exogenous models are better able to explain business cycle fluctuations than is the chaotic endogenous approach. However, our results are subject to a number of caveats, in particular our results could have been influenced by small sample bias, high noise level, incorrect filtering, and long memory of the data.

We investigate a class of estimators for linear regression models where the dependent variable is subject to bid-ask censoring. Our estimation method is based on a definition of error that is zero when the predictor lies between the actual bid price and ask price, and linear outside this range. Our estimator minimizes a sum of such squared errors; it is nonlinear, and indeed the criterion function itself is non-smooth. We establish its asymptotic properties using the approach of Pakes and Pollard (1989). We compare the estimator with midpoint OLS.

We examine the relationship between the risk premium on the S&P500 index total return and its conditional variance. We propose a new semiparametric model in which the conditional variance process is parametric, while the conditional mean is an arbitrary function of the condi-tional variance. For monthly S&P 500 excess returns, the relationship between the two moments that we uncover is nonlinear and nonmonotonic. Moreover, we find considerable persistence in the conditional variance as well as a leverage effect as documented by others.

We introduce a new nonparametric regression estimator that uses prior information on regression shape in the form of a parametric model. In effect, we nonparametrically encompass the parametric model. We obtain estimates of the regression function and its derivatives along with local parameter estimates that can be interpreted from within the parametric model. We establish the uniform consistency and derive the asymptotic distribution of the local parameter estimates and of the corresponding regression and derivative estimates. For estimating the regression function our method has superior performance compared to the usual kernel estimators at or near the parametric model. It is particularly well motivated for binary data using the probit or logit parametric model as a base. We include an application to the Horowitz (1993, Journal of Econometrics 58, 49–70) transport choice dataset.

We examine the relationship between the risk premium on the S&P500 index total return and its conditional variance. We propose a new semiparametric model in which the conditional variance process is parametric, while the conditional mean is an arbitrary function of the conditional variance. For monthly S&P 500 excess returns, the relationship between the two moments that we uncover is nonlinear and nonmonotonic. Moreover, we find considerable persistence in the conditional variance as well as a leverage effect as documented by others.

WE PROPOSE NEW ESTIMATORS of the latent regression function in nonparametric cen- sored and truncated regression models. Our estimators are computationally convenient, consisting only of two nonparametric regressions and a univariate integral. We establish consistency and asymptotic normality for an implementation based on local linear ker- nel estimators. An extension permits estimation in the presence of a general form of heteroscedasticity. Both the function m(.) and the distribution F(.) of the error e are unknown. The errors are not assumed to be symmetric. This paper provides a simple consistent estimator of m(x) + k for some constant k. Under an additional tail condition, we provide a consistent estimator of m(x) when m is the conditional mean function of the uncensored population. Also, we show that F(.) can be estimated given m(x). The proposed estimator is extended to deal with the truncated regression model, where Yi is only observed when it is not censored. We also describe extensions to deal with a general form of heteroskedasticity, in which the distribution of e could depend in unknown ways on a subset of elements of x. For any continuously distributed element Xk of x, let mk(x) = am(x)/axk. We also provide direct estimators of the derivatives mk(x) in both the censored and truncated regression models. These derivatives are interpretable as the marginal effect of a change in x on the underlying uncensored population. They can also be used to test or estimate parametric or semiparametric specifications of m(x). For example, mk(x) is constant if m(x) is linear in Xk, and mk(x) depends only on Xk if m(x) is additive in a function of Xk.

We define new procedures for estimating generalized additive nonparametric regression models that are more efficient than the Linton and Härdle (1996, Biometrika 83, 529–540) integration-based method and achieve certain oracle bounds. We consider criterion functions based on the Linear exponential family, which includes many important special cases. We also consider the extension to multiple parameter models like the gamma distribution and to models for conditional heteroskedasticity.

We stablish the validity of higher order asymptotic expansions to the distribution of a version of the nonlinear semiparametric instrumental variable considered in Newey (1990) as well as to the distribution of a Wald statistic derived from it. We emply local polynomial smoothing with variable bandwidth, which includes local linear, kernel, and [a version of] nearest neighbour estimates as special cases. Our expansions are valid to order n -? for some 0 < ? < ½, where ? depends on the smoothness and dimensionality of the data distribution and on the order of the polynomial chosen by the practitioner. We use the expansions to define optimal bandwidth selection methods for both estimation and testing problems and apply our methods to simulated data.

This paper develops a new estimation procedure for characteristic-based factor models of stock returns. It describes a factor model in which the factor betas are smooth nonlinear functions of observed security characteristics. It develops an estimation procedure that combines nonparametric kernel methods for constructing mimicking portfolios with parametric nonlinear regression to estimate factor returns and factor betas. Factor models are estimated for UK and US common stocks using book-to-price ratio, market capitalizations, and dividend yield.

We examine and compare the finite sample performance of the competing back-fitting and integration methods for estimating additive nonparametric regression using simulated data. Although, the asymptotic properties of the integration estimator, and to some extent the backfitting, method too, are well understood, its small sample properties are not well investigated. Apart from some small experiments in the above cited papers, there is little hard evidence concerning the exact distribution of the estimates. It is our purpose to provide an extensive finite sample comparison between the backfitting procedure and the integration procedure using simulated data.