Identifying and treating outliers in finance

Outliers represent a fundamental challenge in the empirical finance research. We investigate whether the routine techniques used in finance research to identify and treat outliers are appropriate for the data structures we observe in practice. Specifically, we propose a multivariate identification strategy that can effectively detect outliers. We also introduce an estimator that minimizes the bias outliers caused in both cross‐sectional and panel regressions and provide outlier mitigation guidance. Using replications of four recently published studies in premier finance journals, we show how adjusting for multivariate outliers can lead to significantly different results.     

Foreign exchange risk and the predictability of carry trade returns

This paper provides an empirical investigation of the time-series predictive ability of foreign exchange risk measures on the return to the carry trade, a popular investment strategy that borrows in low-interest currencies and lends in high-interest currencies. Using quantile regressions, we find that higher market variance is significantly related to large future carry trade losses, which is consistent with the unwinding of the carry trade in times of high volatility. The decomposition of market variance into average variance and average correlation shows that the predictive power of market variance is primarily due to average variance since average correlation is not significantly related to carry trade returns. Finally, a new version of the carry trade that conditions on market variance generates performance gains net of transaction costs.     

Does the choice of estimator matter when forecasting returns?

While the literature concerned with the predictability of stock returns is huge, surprisingly little is known when it comes to role of the choice of estimator of the predictive regression. Ideally, the choice of estimator should be rooted in the salient features of the data. In case of predictive regressions of returns there are at least three such features; (i) returns are heteroskedastic, (ii) predictors are persistent, and (iii) regression errors are correlated with predictor innovations. In this paper we examine if the accounting of these features in the estimation process has any bearing on our ability to forecast future returns. The results suggest that it does.     

Another Look at the Cross-section of Expected Stock Returns

Our examination of the cross-section of expected returns reveals economically and statistically significant compensation (about 6 to 9 percent per annum) for beta risk when betas are estimated from time-series regressions of annual portfolio returns on the annual return on the equally weighted market index. The relation between book-to-market equity and returns is weaker and less consistent than that in Fama and French (1992). We conjecture that past book-to-market results using COMPUS-TAT data are affected by a selection bias and provide indirect evidence.     

Long-Term Market Overreaction: The Effect of Low-Priced Stocks

Conrad and Kaul (1993) report that most of De Bondt and Thaler's (1985) long-term overreaction findings can be attributed to a combination of bid-ask effects when monthly cumulative average returns (CARs) are used, and price, rather than prior returns. In direct tests, we find little difference in test-period returns whether CARs or buy-and-hold returns are used, and that price has little predictive ability in cross-sectional regressions. The difference in findings between this study and Conrad and Kaul's is primarily due to their statistical methodology. They confound cross-sectional patterns and aggregate time-series mean reversion, and introduce a survivor bias. Their procedures increase the influence of price at the expense of prior returns.     

The long-run relationship between stock prices and goods prices: New evidence from panel cointegration

We examine the long-run relationship between stock prices and goods prices to gauge whether stock market investment can hedge against inflation. Data from 16 OECD countries over the period 1970–2006 are used. We account for different inflation regimes with the use of sub-sample regressions, while maintaining the power of tests in small sample sizes by combining time-series data across our sample countries in a panel unit root and panel cointegration econometric framework. The evidence supports a positive long-run relationship between goods prices and stock prices with the estimated goods price coefficient being in line with the generalized Fisher hypothesis.     

The Determinants of International Commercial Real Estate Investment

We examine the determinants of international commercial real estate investment using a unique set of panel data series for 47 countries worldwide, covering the period from 2000 to 2009. We explore how different socio-economic, demographic and institutional characteristics affect commercial real estate investment activity by determining both cross-sectional and time-series estimators, running augmented random effect panel regressions. We provide evidence that economic growth, rapid urbanization and compelling demographics attract real estate investment, and also demonstrate that a lack of transparency in the legal framework, administrative burdens of doing real estate business, socio-cultural challenges and political instabilities reduce international real estate allocations.     

Forecasting state- and MSA-level housing returns of the US: The role of mortgage default risks

We analyze the ability of an index of mortgage default risks (MDRI) for 43 states and 20 metropolitan statistical areas (MSA) of the US derived from Google search queries, in predicting (in- and out-of-sample) housing returns of the corresponding states and MSAs, based on various panel data and time-series approaches. In general, our results tend to prefer the panel data model based on common correlated effects estimation. We highlight that growth in MDRI negatively impacts housing returns within-sample, with predictive gains primarily concentrated beyond a year. These results are robust to alternative out-of-sample periods and econometric frameworks. Given the role of house prices as a leading indicators, our results are of value to policymakers, especially at the longer-run.     

Jackknifing stock return predictions

We show that the general bias-reducing technique of jackknifing can be successfully applied to stock return predictability regressions. Compared to standard OLS estimation, the jackknifing procedure delivers virtually unbiased estimates with mean squared errors that generally dominate those of the OLS estimates. The jackknifing method is very general, as well as simple to implement, and can be applied to models with multiple predictors and overlapping observations. Unlike most previous work on inference in predictive regressions, no specific assumptions regarding the data generating process for the predictors are required. A set of Monte Carlo experiments show that the method works well in finite samples and the empirical section finds that out-of-sample forecasts based on the jackknife estimates tend to outperform those based on the plain OLS estimates. The improved forecast ability also translates into economically relevant welfare gains for an investor who uses the predictive regression, with jackknife estimates, to time the market.     

Spurious Regressions in Financial Economics?

Even though stock returns are not highly autocorrelated, there is a spurious regression bias in predictive regressions for stock returns related to the classic studies of Yule (1926) and Granger and Newbold (1974) . Data mining for predictor variables interacts with spurious regression bias. The two effects reinforce each other, because more highly persistent series are more likely to be found significant in the search for predictor variables. Our simulations suggest that many of the regressions in the literature, based on individual predictor variables, may be spurious.     

Simple formulas for standard errors that cluster by both firm and time

When estimating finance panel regressions, it is common practice to adjust standard errors for correlation either across firms or across time. These procedures are valid only if the residuals are correlated either across time or across firms, but not across both. This paper shows that it is very easy to calculate standard errors that are robust to simultaneous correlation along two dimensions, such as firms and time. The covariance estimator is equal to the estimator that clusters by firm, plus the estimator that clusters by time, minus the usual heteroskedasticity-robust ordinary least squares (OLS) covariance matrix. Any statistical package with a clustering command can be used to easily calculate these standard errors.     

World market risk,country-specific risk and expected returns in international stock markets

This paper determines whether the world market risk, country-specific total risk, and country-specific idiosyncratic risk are priced in an international capital asset pricing model (ICAPM). Portfolio-level analyses, country-level cross-sectional regressions, stacked time-series, and pooled panel regressions indicate that the world market risk is not, but country-specific total and idiosyncratic risks are significantly priced in an ICAPM framework with partial integration. This result is robust to different methods for estimating risk measures, different investment horizons, and after controlling for the countries’ aggregate dividend yield, earnings-to-price ratios, inflation risk, exchange rate uncertainties, aggregate volatility risk, and past return characteristics. The main findings turn out to be insensitive to the choice of one-factor vs. multifactor models used to estimate systematic and idiosyncratic risk measures.     

Time-varying forecasts by variational approximation of sequential Bayesian inference

Methods developed for making time-varying forecasts in economic and financial analysis include (a) equal-weighted moving-window (or rolling) regression, (b) time-weighted (e.g. exponentially weighted) regression, (c) the Kalman filter (KF) and (d) adaptive Kalman filters. This paper developed a new method based on variational approximation of sequential Bayesian inference (VASB). Concepts and notions of the sequential Bayesian analysis and the variational approximation of an intractable posterior are simple and straightforward. Our VASB algorithm is not complicated and is easy to code. For a regression on multiple time-series, the regression coefficients, standard errors, prediction and residual error are time-varying and are estimated jointly at every time step. For a single time-series (e.g. price returns of an asset), its mean and variance are time-varying and are predicted jointly at every time step. The VASB algorithm performs better than the rolling and time-weighted statistics or regressions and the KF in terms of higher predictive power and stronger robustness. Derivations of the VASB algorithm are presented in the appendices.     

Bank Managers' Heterogeneous Decisions on Discretionary Loan Loss Provisions

This study examines bank managers' three major motivations for discretionary behavior with respect to loan loss provisions: signaling, income smoothing, and capital management. To do so, it utilizes a bank-specific time-series regression approach that captures heterogeneity in the banks' priorities and strategies for alternative motives and compares the results to those from alternative model specifications. The statistical tests and results presented in this study lead to three conclusions. First, significant results for the income smoothing hypothesis are robust to the various model specifications. Second, average signaling coefficients estimated from bank-specific regressions are systematically larger than corresponding coefficients from pooled time-series cross-sectional regressions and are statistically significant. Finally, bank managers appear to use loan loss provisions to manage their regulatory capital levels by comparing them with the minimum ratios specified by regulators rather than with a time-series bank-specific ratio or pooled time-series cross-sectional mean ratio.     

Applied Nonparametric Regression Techniques: Estimating Prepayments on Fixed-Rate Mortgage-Backed Securities

We assess nonparametric kernel-density regression as a technique for estimating mortgage loan prepayments—one of the key components in pricing highly volatile mortgage-backed securities and their derivatives. The highly nonlinear and so-called irrational behavior of the prepayment function lends itself well to an estimator that is free of both functional and distributional assumptions. The technique is shown to exhibit superior out-of-sample predictive ability compared to both proportional-hazards and proprietary-practitioner models. Moreover, the best kernel model provides this improved predictive power utilizing a more parsimonious specification in terms of both data and covariates. We conclude that the technique may prove useful in other financial modeling applications, such as default modeling, and other derivative pricing problems where highly nonlinear relationships and optionality exist.     

Joint Cross-Section/Time-Series Maximum Likelihood Estimation for the Parameters of the Cox-Ingersoll-Ross Bond Pricing Model

We develop a joint maximum likelihood estimator for the interest rate risk premium and the parameters of the Cox, Ingersoll, and Ross bond pricing model. This new approach resolves difficulties inherent in previous approaches that use only time-series data or cross-sectional data. We apply the new approach to a large sample of joint time-series/cross-sectional data. The resulting estimated parameters explain simultaneously the changes in short-term interest rates and the prices of zero-coupon bonds with various maturities. We also identify and provide solutions to potential computational difficulties that other researchers are likely to face.     

Pass-Through of Exchange Rates and Competition between Floaters and Fixers

This paper studies how a rise in the share of U.S. imports from China, or any country with a fixed exchange rate, can explain a disproportionate fall in exchange rate pass-through to U.S. import prices. A theoretical model provides an explanation working through changes in markups, showing that a particular "local bias" condition is necessary and that free entry amplifies the effect. The model produces a structural equation for pass-through regressions including the China share; panel regressions over 1993–2006 indicate that the rising share of trade from China or other exchange rate fixers can explain as much as one-half of the observed decline in pass-through for the United States.     

Stock return,dividend growth and consumption growth predictability across markets and time: Implications for stock price movement

This paper links variation in the predictive regressions for stock returns, dividend growth and consumption growth to economic and market factors. The nature of these links can reveal whether movement in asset prices occurs primarily through the discount rate or cash flow channel, while they also help explain the reported mixed results for predictability. Variation is examined through cross-sectional regressions across 15 markets and over time using rolling regressions. The cross-sectional and time-varying parameters are regressed against output growth, interest rates and inflation as well as market variables using fixed effects panel as well as both OLS and logit approaches. The key implication for asset pricing is that although movement occurs through both channels, stock return predictability is more dominated by the discount rate channel and consumption growth predictability more so by the cash flow channel. Intuitively, such a difference may arise as investors and households rebalance their asset holdings and consumption at different speeds. There is also some evidence of money illusion through the inflation variable.     

Home Bias in Emerging Bond and Stock Markets

In this paper, we empirically examine sizes and sources of home bias in both bond and equity markets for twenty emerging countries and twenty-two developed countries over the 2001-11 sample period. The average size of home bias in both bond and stock markets is found to be much larger in emerging countries than in developed countries. Using the explanatory variables in two categories of economic development and market performance, we employ dynamic panel data regression models to analyze major sources of home bias. The main results are the following: First, market performance factors generally affect home bias more strongly than do economic development factors. Second, market factors including market return, volatility, and liquidity support various hypotheses under informational asymmetries, such as return chasing, risk aversion, and flight to quality. Third, among macroeconomic factors, it is shown that real gross domestic product growth has negative effects and country leverage has positive effects on a specific home bias, backing up the size-bias and the flight-to-quality hypotheses, respectively. Finally, and perhaps most important in this paper, the effect of bond market performance on equity home bias is found to be significantly stronger than the effect of equity market performance on bond home bias from the market interaction model estimation, suggesting that a policy design needs to begin with increasing bond market efficiency to reduce equity market home bias.