Francesco Lisi

Is a random walk the best exchange rate predictor

Abstract The paper discusses short-term exchange rate prediction, using the random walk hypothesis (RWH) as a benchmark to compare performances. After surveying some recent results in this field, the authors suggest a filtering-prediction method inspired by recent developments in nonlinear dynamical systems theory. The filtering of presumably noisy data is realized by means of a technique derived from Singular Spectrum Analysis (SSA) conveniently adapted to a nonlinear dynamics context. In particular, the authors develop a multichannel version of SSA. Filtered data are then used to perform an out-of-sample, short-term prediction, by means of a nonlinear (locally linear) method. This method is applied to exchange rate series of the major currencies and the predictions thus obtained are shown to outperform neatly those derived from the RWH. Finally, the application of a test recently developed by Mizrach confirms the statistical significance of the results.

Combining Singular-Spectrum Analysis and neural networks for time series forecasting

In this paper, we propose a combination of an adaptive noise-reduction algorithm based on Singular-Spectrum Analysis (SSA) and a standard feedforward neural prediction model. We test the forecast skill of our method on some short real-world and computergenerated time series with different amounts of additive noise. The results show that our combined technique has better performances than those offered by the same network directly applied to raw data, and therefore is well suited to forecast short and noisy time series with an underlying deterministic data generating process (DGP).

Looking for Skewness in Financial Time Series

In this paper, we study marginal and conditional skewness in financial returns for nine time series of major international stock indices. For this purpose, we develop a new variant of the GARCH model with dynamic skewness and kurtosis. Our empirical results indicate that there is no evidence of marginal asymmetry in the nine time series under consideration. We do however find significant time-varying conditional skewness. The economic significance of conditional skewness is analysed in terms of Value-at-Risk measures and Market Risk Capital Requirements set by the Basel Accord. Copyright

Periodic Long-Memory GARCH Models

A distinguishing feature of the intraday time-varying volatility of financial time series is given by the presence of long-range dependence of periodic type, due mainly to time-of-the-day phenomena. In this work, we introduce a model able to describe the empirical evidence given by this periodic long-memory behaviour. The model, named PLM-GARCH (Periodic Long-Memory GARCH), represents a natural extension of the FIGARCH model proposed for modelling long-range persistence of volatility. Periodic long memory versions of EGARCH (PLM-EGARCH) and of Log-GARCH (PLM-LGARCH) models are also examined. Some properties and characteristics of the models are given and finite sample performance of quasi-maximum likelihood estimation are studied with Monte Carlo simulations. Further possible extensions of the model to take into account multiple sources of periodic long-memory behaviour are proposed. Two empirical applications on intra-day financial time series are also provided.

Generalised long-memory GARCH models for intra-daily volatility

The class of fractionally integrated generalised autoregressive conditional heteroskedastic (FIGARCH) models is extended for modelling the periodic long-range dependence typically shown by volatility of most intra-daily financial returns. The proposed class of models introduces generalised periodic long-memory filters, based on Gegenbauer polynomials, into the equation describing the time-varying volatility of standard GARCH models. A fitting procedure is illustrated and its performance is evaluated by means of Monte Carlo simulations. The effectiveness of these models in describing periodic long-memory volatility patterns is shown through an empirical application to the Euro-Dollar intra-daily exchange rate.

Practical implications of higher moments in risk management

In this paper the out-of-sample prediction of Value-at-Risk by means of models accounting for higher moments is studied. We consider models differing in terms of skewness and kurtosis and, in particular, the GARCHDSK model, which allows for constant and dynamic skewness and kurtosis. The issue of VaR prediction performance is approached first from a purely statistical viewpoint, studying the properties concerning correct coverage rates and independence of VaR violations. Then, financial implications of different VaR models, in terms of market risk capital requirements, as defined by the Basel Accord, are considered. Our results, based on the analysis of eight international stock indexes, highlight the presence of conditional skewness and kurtosis, in some case time-varying, and point out that asymmetry plays a significant role in risk management.

Testing asymmetry in financial time series

This paper examines the problem of evaluating the presence of asymmetry in the marginal distribution of financial returns by means of a suitable statistical test. After a brief description of existing tests, a bootstrap procedure is proposed. A Monte Carlo study showed that this test works properly and that, in terms of power, it is competitive with existing tests. An application to real financial time series is also presented.

k -Factor GARMA models for intraday volatility forecasting

This paper studies the ability of the k -factor GARMA processes to model and forecast the volatility of an intraday financial time series. Forecasting results from the k -factor GARMA model are obtained and compared with those produced by a conventional SARIMA model.

Predictive accuracy for chaotic economic models

Abstract In this work we present a technique to obtain prediction intervals for chaotic data. Using nearest neighbors method we give estimates of local variance and percentiles of the prediction error distribution. This allows to define an interval containing a future value with a given probability. Its effectiveness is shown with data generated by a chaotic economic model.

Nonlinear models for ground--level ozone forecasting

One of the main concerns in air pollution is excessive tropospheric ozone concentration. The aim of this work is to develop statistical models giving shortterm forecasts of future ground-level ozone concentrations. Since there are few physical insights about the dynamic relationship between ozone, precursor emissions and/or meteorological factors, a nonparametric and nonlinear approach seems promising in order to specify the forecast models. First, we apply four nonparametric procedures to forecast daily maximum 1-hour and maximum 8-hour averages of ozone concentrations in an urban area. Then, in order to improve the forecast performances, we combine the time series of the forecasts. This idea seems to give encouraging results.