Nonlinear Determinants of Unemployment in Türkiye: A MARS Approach With Scale-Invariant Model Selection

Zenodo formundaki alanlara doğrudan yapıştırabileceğin metinler aşağıda. Upload type: Dataset. Paket içinde kod da olduğu için "Software" da seçilebilir, ama Dataset daha uygun. Title: Replication package for "Scale-Invariant Information Complexity and Knot-Cost-Aware Criteria for MARS Model Selection: Monte Carlo Evidence and a Time-Series Application to Unemployment in Türkiye" Authors: Meryem Bekar Adıgüzel (Aksaray University) – Mehmet Ali Cengiz (Imam Mohammad Ibn Saud Islamic University) Description: This repository contains the data, code, tables and figures needed to reproduce all results reported in the article "Scale-Invariant Information Complexity and Knot-Cost-Aware Criteria for MARS Model Selection: Monte Carlo Evidence and a Time-Series Application to Unemployment in Türkiye" (Bekar Adıgüzel and Cengiz, submitted to Heliyon). The study compares model-selection criteria for multivariate adaptive regression splines (MARS) on a common backward-pruning path. The criteria are generalized cross-validation with different knot charges, five-fold cross-validation, AIC, AICc, SBC, ICOMP(IFIM) and ICOMP(IFIM)_PEU. The study also proposes knot-cost-aware likelihood criteria (AIC_eff, SBC_eff) and a scale-invariant information-complexity criterion (ICOMP_PEU-S). The criteria are evaluated in a Monte Carlo study with 12 designs and 200 replications per design (2,400 data sets). They are then applied to quarterly unemployment in Türkiye (2015Q1–2025Q2) using a level specification and a time-series-consistent specification. The application includes unit-root tests, residual diagnostics, rolling-origin forecasts with Diebold–Mariano tests, and moving-block bootstrap stability analysis. Contents data/: quarterly macroeconomic data for Türkiye, 2015Q1–2025Q2 (n = 42): unemployment rate, industrial production index, tertiary-education graduates, CPI inflation, interest rate, USD/TRY exchange rate, imports and exports. Provided as CSV and as an Excel workbook in which the level and time-series designs are built with formulas. The data were compiled from the Turkish Statistical Institute (TURKSTAT) and the Central Bank of the Republic of Türkiye (CBRT). R/: R and Python scripts: the selection-criteria engine, the Monte Carlo study, the empirical analysis, the scale-invariance experiment, forecasting and figures. out/tab/: all numerical results as CSV files, including the raw Monte Carlo replications. out/fig_paper/: all figures of the article (PNG and PDF, 300 dpi). tables/: all tables of the article (Tables 1–12 and A1) and supplementary Tables S1–S8 in a single Excel workbook. paper/: manuscript sources and build scripts. README.md and run_all.sh: instructions and a single script that reproduces every result. Software: R 4.3.3 (earth 5.3.3, forecast 8.21.1, tseries 0.10-55, lmtest, sandwich, MASS) and Python 3.11 (pandas, numpy, matplotlib, statsmodels, openpyxl). Random seeds are fixed in all scripts.