Answers are grounded in this paper and cite their sources.
This study assesses machine learning’s incremental forecasting value for the Borsa Istanbul Tourism Index through directional, probabilistic, and intraday economic evaluations. Twenty-three market-based predictors from daily 2010–2026 data were used at 1644 forecast origins in 2020–July 2026. Hyperparameters were selected using outcomes realized before 2020, and models were refitted quarterly. Logistic regression, random forest, XGBoost, and long short-term memory networks were compared with majority, historical-prevalence, directional-persistence, and first-order probabilistic-persistence benchmarks. Paired circular moving-block bootstrap inference addressed serial dependence and multiple comparisons. Inference was conditional on selected configurations and fitted models, excluding uncertainty from repeated hyperparameter selection and model fitting. Directional persistence achieved the highest Matthews correlation coefficient (MCC), 0.110, versus approximately 0.088 for logistic regression and random forest. No fitted model established MCC superiority over persistence. XGBoost’s Brier score was 0.24610, versus 0.24888 for historical prevalence and 0.24738 for probabilistic persistence; primary adjusted comparisons did not establish superiority over either benchmark. Separately selected intraday configurations aligned forecasts with holding periods. Economic outcomes varied across periods, with uncertain excess returns over interest-bearing cash. Distributed outputs support numerical verification; full reproduction requires authorized source data and the documented estimation environment.
No takes yet. Share an insight, caveat, or question.
Sakınç et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: