Title: Revealing technical trading rules with the empirical similarity concept
Authors: Yarema Okhrin - Universitaet Augsburg (Germany) [presenting]
Abstract: Economists frequently suggest formal mathematical models or theories that postulate a formal decision rule. These rules can be subsequently calibrated and validated using empirical data. When there is no suitable rule at hand, economic agents may rely on so-called case-based inference, which grounds on analyzing cases by drawing analogies between experienced situations and their outcomes and their analogy to the present problem. The empirical similarity concept puts the case-based decision theory into an econometric framework, by predicting the variable of interest by a sum of historical outcomes weighted by the distances between the current levels of covariates and their historical counterparts. We formalize the tools of technical analysis aimed to forecast asset prices and returns. In contrary to mostly heuristic definition of patterns in technical analysis, we rely on the B-spline regularization to define and classify the patterns. This technique can also be seen as a decomposition of asset prices and a parametric quantification of patterns typical in technical analysis. The empirical application shows that the obtained patters deviate from the patterns commonly assumed in practice. Thereafter, we apply empirical similarity to compare the historical patters with the current one and to build a forecast for the future price/return dynamics. We compare the approach to common benchmarks.