Jan Antell

Testing for cointegration between international stock prices

This paper re-examines the evidence for cointegration between international stock prices. It applies Johansens maximum likelihood (ML) cointegration method and likelihood ratio (LR) tests for cointegration to stock prices. In monthly data it finds at most one cointegrating vector and in quarterly data finds no cointegrating vectors. Using the small-sample corrections or the small-sample critical values it finds no evidence of cointegration. Johansens LR tests for cointegration are sensitive to the lag length specification in the VAR model. In general it finds more evidence of cointegration in higher order VAR models. The paper shows that some of the previous empirical results can be explained by the small-sample bias and size distortion of Johansens LR tests for cointegration. It finds that international stock prices are not cointegrated.

Bootstrap and fast double bootstrap tests of cointegration rank with financial time series

The likelihood ratio test of cointegration rank is the most widely used test for cointegration. Many studies have shown that its finite sample distribution is not well approximated by the limiting distribution. Bootstrap and fast double bootstrap (FDB) algorithms for the likelihood ratio test are introduced and evaluated by Monte Carlo simulation experiments. It is found that the performance of the ordinary (single) bootstrap test is in most cases good in terms of the size of the test. The FDB produces a further improvement in cases where the performance of the asymptotic test is unsatisfactory and the single bootstrap test overrejects noticeably. The FDB is shown to be a useful supplement to the single bootstrap as a tool for determining the cointegration rank. The tests are applied to US interest rates and international stock prices series. By simulating the data assuming that the cointegration rank is known, it is found that the asymptotic test tends to overestimate the cointegration rank, while the bootstrap and FDB tests choose the correct cointegration rank.

Countercyclical and Time-Varying Risk Aversion and Equity Premium

This paper tests the counter-cyclicality of aggregate risk aversion and price of market risk using a novel testing approach introduced in Antell and Vaihekoski (2015) for conditional asset pricing models. Cohen et al. (2015) report experimental evidence that the risk aversion is countercyclical, although empirical support from financial studies is at best inconclusive. This paper applies the new testing approach for the Merton (1973, 1980) model with time-varying risk aversion. The testable implications link realized equity premium to, among others, changes in conditional variance, its long-term persistence, and changes in the time-varying risk aversion. Empirically, testing is conducted using monthly US stock market data from 1928 to 2013, and using (asymmetric) GARCH models to estimate conditional variance. We compare various methods to model economic expectations regarding the state of the economy. Unlike the traditional estimation approach, the results from the new estimation approach give support for time-varying and countercyclical behavior for the risk aversion.

Expected Return and Conditional Asset Pricing: A New Testing Approach

Traditional tests of conditional asset pricing models are performed under the assumption of rational expectations and presume that the use of realized returns as a proxy for expected returns is acceptable. This paper turns the tables and asks what realized returns we would observe, given the pricing model. Based on this idea, we develop a new approach for testing conditional asset pricing models. The new approach implies a testable model that links the realized returns to changes in the risk-free rates, expected dividends and risk premium. This approach is used to test the Merton (1973, 1980) model and a long-standing risk-return trade-off puzzle: the price of market risk, lambda, has often turned out to be small, even negative and often insignificant. The empirical model implied by the new approach suggests that estimating lambda requires that we take into account changes in the conditional variance and its long-term persistence. We compare the price of market risk estimates from the new and the traditional testing approaches. We use both conventional measures of variance, such as (asymmetric) GARCH estimates as well as new measures based on forward-looking option market implied volatilities and MIDAS estimation. Tests are conducted using US stock market data. The re-sults give strong support for the new testing approach; the lambda estimates from the new approach are consistently significant, positive, more stable, and higher than those estimated using the traditional approach.