This paper introduces a time-varying threshold autoregressive model (TVTAR), which is used to examine the persistence of deviations from PPP. We find support for the stationary TVTAR against the unit root hypothesis; however, for some developing countries, we do not reject the TVTAR with a unit root in the corridor regime. We calculate magnitudes, frequencies, and durations of the deviations of exchange rates from forecasted changes in exchange rates. A key result is asymmetric adjustment. In developing countries, the average cumulative deviation from forecasts during periods when exchange rates are below forecasts is twice the corresponding measure during periods when exchange rates are above forecasts.

In the last two years or so, I was most fortunate in being given opportunities of lecturing on a new methodology to a variety of audiences in Britain, China, Finland, France and Spain. Despite my almost Confucian attitude of preferring talking (i.e. a transient record) to writing (i.e. a permanent record), the warm encouragement of friends has led to the ensuing notes. I am also only too conscious of the infancy of the methodology introduced in these notes. However, it is my sincere hope that exposure to a wider audience will accelerate its maturity. Readers are assumed to be familiar with the basic theory of time series analysis. The book by Professor M.B. Priestley (1981) may be used as a general reference. Chapter One is addressed to the general question: "why do we need non-linear time series models?" After describing some significant advantages of linear models, it singles out several major limitations of linearity. Of course, the selection reflects my personal view on the subject, which is only at its very beginning, although there does seem to be a general agreement in the literature that time irr'eversibility and limit cycles are among the most obvious.

This book constitutes the refereed proceedings of the joint conference on Machine Learning and Knowledge Discovery in Databases: ECML PKDD 2010, held in Barcelona, Spain, in September 2010. The 120 revised full papers presented in three volumes, together with 12 demos (out of 24 submitted demos), were carefully reviewed and selected from 658 paper submissions. In addition, 7 ML and 7 DM papers were distinguished by the program chairs on the basis of their exceptional scientific quality and high impact on the field. The conference intends to provide an international forum for the discussion of the latest high quality research results in all areas related to machine learning and knowledge discovery in databases. A topic widely explored from both ML and DM perspectives was graphs, with motivations ranging from molecular chemistry to social networks.