Adaptive machine learning in stock markets is an important system for reacting to new information in real-time, automatic prediction and completing tasks independently, using suitable algorithms. Professor Duran talked about their several studies toward this goal at Koç University on the 12th of November 2024. Duran and Caginalp [1] propose a hybrid parameter optimization forecast algorithm including daily based learning with two streaming windows, using a semi-dynamic initial parameter vector pool having not only fixed but also most recently used successful parameter vectors from a set of grid points in a hyperbox and out-of-sample prediction. More recently, Tuncel and Duran [2] propose a new mathematical method and focus on Monte Carlo simulation to find out the effectiveness of two approaches, including the grid approach and the random approach in hyperbox based on the experimental design for a selection of initial parameter vectors in a large-scale unconstrained optimization problem. Moreover, Duran and Bommarito [3] present a new profitable trading and risk management strategy with transaction cost for an adaptive, equally weighted portfolio. It was elected as a key paper in risk by the Quantitative Finance Journal in 2010.

[1] Ahmet Duran and Gunduz Caginalp, Parameter optimization for differential equations in-asset price forecasting, Optimization Methods & Software 23 (4) (2008) 551–574, Issue: Mathematical programming in data mining and machine learning.

[2] Mehmet Tunçel and Ahmet Duran, Effectiveness of grid and random approaches for a model parameter vector optimization, Journal of Computational Science 67 (2023) 101960

[3] Ahmet Duran and Michael J. Bommarito, A profitable trading and risk management strategy despite transaction costs, Quantitative Finance, 11(6), 2011, 829-848