Research

Interested in time series analysis, macro and financial econometrics.

Especially noninvertibility and non-causality in time series models, and statistical models with heavy tails. Also, structural VARMA models and applications in empirical macro and finance.

Thesis supervisor: associate professor Mika Meitz.


Working Papers (latest versions):

  • Residual-based diagnostic tests for noninvertible ARMA models (R&R JTSA)
    • This paper proposes two residual-based diagnostic tests for noninvertible ARMA models. The tests are analogous to the portmanteau tests developed by Box and Pierce (1970) and McLeod and Li (1983) in the conventional invertible case. We derive the asymptotic χ-squared distribution for the tests under the null of correctly specified model, and study the size and power properties in a Monte Carlo simulation study. An empirical application employing financial time series data points out the usefulness of noninvertible ARMA model in analyzing stock returns and the use of the proposed test statistics.

  • Maximum likelihood estimation of a noninvertible ARMA model with α-stable errors
    • We study properties of the maximum likelihood estimator of in a noninvertible ARMA model with errors that follow an α-stable distribution and have infinite variance. To ensure the identification of the noninvertible ARMA model considered, we restrict the analysis to non-Gaussian distributions. Estimators of the autoregressive and moving average parameters are shown to be n^1/α-consistent and to converge to a non-standard limiting distribution that is obtained as a maximizer of a certain random function. Estimators of the parameters in the α-stable distribution have the conventional n^1/2 rate of convergence. Finite sample properties of the estimators are studied in a simulation experiment, and an application to financial time series data from the New York Stock Exchange illustrates the applicability of the model.

  • Nonlinear predictability of asset returns
    • For many theoretical asset pricing models, predictability follows as an implication of risk aversion of agents. A closed form solutions for the next periods asset return depends on how the agents form their expectations about the future state of the world. By no means should this predictability be linear. First, we provide evidence of predictability of returns of U.S. stock portfolios and individual financial sector stocks using noninvertible ARMA model and two-stage predictability testing procedure by Lanne, Meitz, and Saikkonen (2013). Second, we provide a straightforward extensions to this procedure and allow for a larger model than noninvertible ARMA(1,1).