Forecasting Financial Volatility Using Nested Monte Carlo Expression Discovery

Abstract

We are interested in discovering expressions for financial prediction using Nested Monte Carlo Search and Genetic Programming. Both methods are applied to learn from financial time series to generate non linear functions for market volatility prediction. The input data, that is a series of daily prices of European S&P500 index, is filtered and sampled in order to improve the training process. Using some assessment metrics, the best generated models given by both approaches for each training sub sample, are evaluated and compared. Results show that Nested Monte Carlo is able to generate better forecasting models than Genetic Programming for the majority of learning samples.

Adaptive General Search Framework for Games and Beyond

“…Various applications of search algorithms can be found also in logistic [3] and robotics [65], [66]. Many more examples can be mentioned, including structural engineering [6], protecting natural resources from illegal extraction [67], forecasting financial volatility of assets [68], managing wildfires [69], improving computeraided retrosynthesis [70] and maximizing the performance of job scheduling heuristics [71]. Many more applications are being explored every day.…”

Stabilized Nested Rollout Policy Adaptation

“…The Adapt Algorithm is given in algorithm 2.For all the states of the sequence passed as a parameter it adds α to the weight of the move of the sequence (lines 3-5). Then it reduces all the moves proportionally to α times the probability of playing the move so as to keep a sum of all probabilities equal to one (lines [6][7][8][9][10][11][12].…”

“…At greater levels it performs N iterations and for each iteration it calls itself recursively to get a score and a sequence (lines 4-7). If it finds a new best sequence for the level it keeps it as the best sequence (lines [8][9][10][11]. Then it adapts the policy using the best sequence found so far at the current level (line 12).…”

“…Nested Monte Carlo Search can be used for the optimization of mathematical expressions [5,6,8]. For some problems it gives better results than alternative algorithms such as UCT [17] or Genetic Programming [18].…”

Monte Carlo Tree Search in the Presence of Transition Uncertainty

“…It manages the balance of exploration and exploitation with techniques such as UCT (Kocsis, Szepesvári, and Willemson 2006). Often combined with machine learning, it has been enormously successful in both games (Silver et al 2016;Gao, Müller, and Hayward 2018;Gao 2020;Saffidine 2008;Nijssen and Winands 2010) and non-game applications (Lu et al 2016;Mansley, Weinstein, and Littman 2011;Sabharwal, Samulowitz, and Reddy 2012;Cazenave 2010). In these applications, a perfect simulation model allows for efficient lookahead search.…”