Search

Performance of mutual fund selection methods is typically assessed using long samples (long time series). It is, however, very often of interest how well the methods perform in shorter samples. We carry out an extensive simulation study based on an empirically motivated skill distribution. For both short and long samples, we present evidence of large differences in performance between popular fund selection methods. In an empirical analysis, we show that the differences documented by the simulations are empirically relevant.

There exists an extensive literature estimating idiosyncratic labor income processes. While a wide variety of models are estimated, GMM estimators are almost always used. We examine the validity of using likelihood based estimation in this context by comparing the small sample properties of a Bayesian estimator to those of GMM. Our baseline studies estimators of a commonly used simple earnings process. We extend our analysis to more complex environments, allowing for real world phenomena such as time varying and heterogeneous parameters, missing data, unbalanced panels, and non-normal errors. The Bayesian estimators are demonstrated to have favorable bias and efficiency properties.

We describe exact inference based on group-invariance assumptions that specify various forms of symmetry in the distribution of a disturbance vector in a general nonlinear model. It is shown that such mild assumptions can be equivalently formulated in terms of exact confidence sets for the parameters of the functional form. When applied to the linear model, this exact inference provides a unified approach to a variety of parametric and distribution-free tests. In particular, we consider exact instrumental variable inference, based on symmetry assumptions. The unboundedness of exact confidence sets is related to the power to reject a hypothesis of underidentification. In a multivariate instrumental variables context, generalizations of Anderson-Rubin confidence sets are considered. (C) 2007 Elsevier B.V. All rights reserved.

The main purpose of the analysis of the cointegrated VAR model is conducting inference on the cointegrating relations. Asymptotic inference is chi(2), but the asymptotic results are not accurate enough for small samples. Therefore, we derive here a correction factor, depending on sample size and parameters, for the likelihood ratio test of some linear hypotheses on the cointegrating space in a vector autoregressive model. We have to assume that the adjustment coefficients are known. The main idea is to condition on the common trends when calculating the correction factor. Some simulation experiments illustrate the findings. The article is a published version of EUI ECO WP; 1999/09

In this paper, we compare the properties of the main criteria proposed for selecting the number of factors in dynamic factor model in a small sample. Both static and dynamic factor numbers' selection rules are studied. Simulations show that the GR ratio proposed by Ahn and Horenstein (2013) and the criterion proposed by Onatski (2010) outperform the others. Furthermore, the two criteria can select accurately the number of static factors in a dynamic factors design. Also, the criteria proposed by Hallin and Liska (2007) and Breitung and Pigorsch (2009) correctly select the number of dynamic factors in most cases. However, empirical applications show most criteria select only one factor in presence of one strong factor.

The authors first discuss performance evaluation using stochastic discount factors and relate it to traditional mean-variance analysis. They then use Monte Carlo experiments to examine the properties of various general method of moment (GMM) estimators. The test statistics are fairly well behaved although serious size distortions are found in some cases. The simulations also show that a significant excess return, or a long sample, is needed to reject neutral performance. Finally, the authors offer an evaluation of Swedish-based mutual funds. The conditional evaluation indicates that funds have had nonneutral performance as revealed by the predictability of the unconditional performance measure. Copyright 1999 by University of Chicago Press.

In this paper we examine the possibility of using the standard Kruskal-Wallisrank test in order to evaluate whether the distribution of efficiency scores resulting from Data Envelopment Analysis (DEA) is independent of the input(or output) mix. Recently, a general data generating process (DGP) suiting the DEA methodology has been formulated and some asymptotic properties of the DEA estimators have been established. In line with this generally accepted DGP, we formulate a conditional test for the assumption of mix independence. Since the DEA frontier is estimated, many standardl assumptions for evaluating the test statistic are violated. Therefore, we propose to explore its statistical properties by the use of simulation studies. The simulations are performed conditional on the observed input mixes. The method, as shown here, is applicable for models with multiple inputs and one output with constant returns to scale when comparing distributions of efficiency scores in two or more groups. The approach is illustrated in an empirical case of demolition projects where we reject the assumption of mix independence. This means that it is not meaningful to perform a complete ranking of the projects based on their efficiency score. Thus the example illustrates how common practice can beinappropriate.

Copula-GARCH models have been recently proposed in the financial literature as a statistical tool to build flexible multivariate distributions. Our extensive simulation studies investigate the small sample properties of these models and examine how misspecification in the marginals may affect the estimation of the dependence function represented by the copula. We show that the use of normal marginals when the true Data Generating Process is leptokurtic or asymmetric, produces negatively biased estimates of the normal copula correlations. A striking result is that these biases reach their highest value when correlations are strongly negative, and viceversa. This result remains unchanged with both positively skewed and negatively skewed data, while no biases are found if the variables are uncorrelated. Besides, the effect of marginals asymmetry on correlations is smaller than that of leptokurtosis. We finally analyse the performance of these models in terms of numerical convergence and positive definiteness of the estimated copula correlation matrix.

This paper sets out to help explain why estimates of asset correlations based on equity prices tend to be considerably higher than estimates based on default rates. Resolving this empirical puzzle is highly important because, firstly, asset correlations are a key driver of credit risk and, secondly, both data sources are widely used to calibrate risk models of financial institutions. By means of a simulation study, we explore the hypothesis that differences in the correlation estimates are due to a substantial downward bias characteristic of estimates based on default rates. Our results suggest that correlation estimates from equity returns are more efficient than those from default rates. This finding still holds if the model is misspecified such that asset correlations follow a Vasicek process which affects foremost the estimates from equity returns. The results lend support for the hypothesis that the downward bias of default-rate based estimates is an important although not the only factor to explain the differences in correlation estimates. Furthermore, our results help to quantify the estimation error of asset correlations dependent on the risk characteristics of the underlying data base. Abhängigkeiten zwischen den Ausfallereignissen von Kreditnehmern sind ein wesentlicher Treiber des Kreditrisikos in Kreditportfolien. Solche Abhängigkeiten werden gewöhnlich durch Asset-Korrelationen zwischen Firmenwertänderungen gemessen. Da Firmenwertänderungen nicht beobachtbar sind, werden diese Korrelationen oft aus Zeitreihen von Aktienrenditen oder aus historischen Ausfallraten geschätzt. Beide Ansätze haben in der Forschung zu erheblich unterschiedlichen Ergebnissen geführt. Da empirische Untersuchungen unterschiedliche Stichproben verwenden, ist es bisher nicht möglich gewesen, diese Unterschiede zu erklären. In diesem Arbeitspapier untersuchen wir die Hypothese, dass die beobachteten Unterschiede sich aus unterschiedlichen statistischen Eigenschaften der Schätzmethoden erklären, die jeweils bei der Schätzung aus Aktienrenditen und aus Ausfallraten verwendet werden. Eine Bestätigung der Hypothese kann Kreditrisikomanagern eine Hilfestellung geben bei der Auswahl der geeigneten Datenquelle für die Schätzung von Asset-Korrelationen. Um diese Hypothese zu bestätigen, verwenden wir eine umfassende Simulationsstudie mit einer Vielzahl von Risikoparametern und unterschiedlich großen Kreditportfolien. Wir beobachten, dass die statistischen Methoden eine wichtige Rolle bei der Erklärung der Unterschiede zwischen den Schätzwerten von Asset-Korrelationen basierend auf Aktienrenditen oder Ausfallraten spielen. Es ist grundsätzlich empfehlenswert, Aktienrenditen für die Schätzung zu verwenden, da die statistischen Fehler in diesem Fall geringer sind. Diese Beobachtung gilt auch, falls das Modell insofern fehlspezifiziert ist, als die Asset-Korrelationen nicht wie im Model unterstellt über die Zeit konstant sind, sondern einem stochastischen Prozess folgen.

This paper proposes new sets of moment restrictions for consistently estimating the dynamic panel data model. These sets are derived from solving the moment generating functions of the error term for the dynamic panel data model and have the relevancy with some well-known sets of moment restrictions proposed up to this point in time. To investigate small sample properties for GMM estimators based on these sets, Monte Carlo experiments were conducted. The Monte Carlo experiments show that the GMM estimators based on some of these sets exhibit good small sample properties for some values of the so-called adjusting parameter.

In the thesis we consider inference for cointegration in vector autoregressive (VAR) models. The thesis consists of an introduction and four papers. The first paper proposes a new test for cointegration in VAR models that is directly based on the eigenvalues of the least squares (LS) estimate of the autoregressive matrix. In the second paper we compare a small sample correction for the likelihood ratio (LR) test of cointegrating rank and the bootstrap. The simulation experiments show that the bootstrap works very well in practice and dominates the correction factor. The tests are applied to international stock prices data, and the .nite sample performance of the tests are investigated by simulating the data. The third paper studies the demand for money in Sweden 1970—2000 using the I(2) model. In the fourth paper we re-examine the evidence of cointegration between international stock prices. The paper shows that some of the previous empirical results can be explained by the small-sample bias and size distortion of Johansen’s LR tests for cointegration. In all papers we work with two data sets. The first data set is a Swedish money demand data set with observations on the money stock, the consumer price index, gross domestic product (GDP), the short-term interest rate and the long-term interest rate. The data are quarterly and the sample period is 1970(1)—2000(1). The second data set consists of month-end stock market index observations for Finland, France, Germany, Sweden, the United Kingdom and the United States from 1980(1) to 1997(2). Both data sets are typical of the sample sizes encountered in economic data, and the applications illustrate the usefulness of the models and tests discussed in the thesis.

The small sample properties of the systemwise RESET test for functional misspecification is investigated using normal and non-normal error terms. When using normally distributed or less heavy tailed error terms, we find the Rao's multivariate F-test to be best among all other alternative test methods. Using the bootstrap critical values, however, all test methods perform satisfactorally in almost all situations. However, the test methods perform extremely badly (even the RAG test) when the error terms are very heavy tailed.

The performance of the FPE, AIC, HQ and SC criteria in choosing lag-length, and the effect on the impulse-response functions, are studied in a Monte Carlo simulation. The experiments include stationary, cointegrated, and mixed unit root VAR and MA cases.