Topik Terkait
Forecasting, Keuangan, Machine Learning
- This study develops a practical forecast for IndONIA 1-month at quarter end and identifies the forces that shape its path. A harmonized quarterly dataset from CEIC and Bloomberg underpins the analysis, with series aligned to common timestamps, units, and currency conventions and checked against source releases.
- The baseline uses SARIMAX and Spread Analysis to map observable policy, currency, and short-tenor market information into an interpretable forecast. Machine learning and deep learning models including Elastic Net, Gradient Boosting, Random Forest, SVR, XGBoost, CatBoost, LSTM, and GRU serve as cross checks on direction, turning points, and relative magnitudes rather than substitutes for the baseline.
- Results are straightforward. Domestic policy settings and external policy benchmarks load positively on quarter-end IndONIA, while currency conditions and very short money market measures provide timely confirmation and often lead shifts in the backdrop. Equity market moves add limited incremental information once policy and currency variables are included. Out of sample tests show that the baseline captures direction and turning points with errors that are small enough for planning, and backtests track the recent decline credibly.
- Consequently, for insurance balance sheets the baseline forecast, and ML/DL cross-checks translate directly into pricing and crediting-rate guidance, ALM rebalancing and duration positioning, liquidity buffers around policy dates, and reserve sensitivity to policy and currency shocks, thereby supporting RBC compliance and stress testing.
- The framework is designed for quarterly horizons where policy and near-term market signals carry most of the weight. Very abrupt intra quarter shocks can still move outcomes between forecast points, so interim monitoring remains important. Full specifications, variable definitions, and hyperparameters are provided in the Appendix.
