Forecasting the USD/PHP Exchange Rate

Time-series forecasting of the USD/PHP daily exchange rate over six fiscal years (FY2020–FY2026) using an ARIMA(0,1,1)–GARCH(1,1) ensemble alongside deep-learning baselines (Simple RNN, LSTM, Transformer, 1D-CNN), benchmarked against a naive-persistence baseline.

Final project for DS1321 — Time Series Analysis and Stochastic Processes, BS Data Science, University of Asia and the Pacific. Group of 2 (David Riego + 1 partner).

Repository layout

notebooks/        Jupyter notebooks — EDA, ARIMA-GARCH, Deep Learning forecasting
data/             Train / validation / test CSV splits + cleaned dataset
figures/          All generated plots (ACF/PACF, seasonal decomposition, model
                  forecasts, baselines vs optimal, etc.)
reports/          Final paper, EDA documentation, professor's feedback, slides,
                  earlier proposal versions
presentation/     Deep Learning for Time Series slide deck (PDF)
medium-article/   Public-facing Medium-style technical write-up (HTML + PDF)
interactive/      Interactive Plotly HTML — forecast bands, GARCH volatility
misc/             Auxiliary spreadsheets

Methods

1. EDA & Preprocessing — Log-return transformation; seasonal decomposition framed by fiscal year; train/val/test splits.
2. Classical: ARIMA(0,1,1) for the mean process + GARCH(1,1) for conditional heteroskedasticity. Multi-horizon out-of-sample evaluation.
3. Deep learning: Simple RNN, LSTM, Transformer, 1D-CNN — all benchmarked against naive persistence.
4. Evaluation: Out-of-sample RMSE / MAE on the held-out FY2024–FY2026 set; interactive forecast-band visualization.

Stack

Python · pandas · numpy · statsmodels (ARIMA, GARCH) · PyTorch (RNN, LSTM, Transformer, CNN) · matplotlib · plotly

Status

Course final project, completed Mar–May 2026. Not deployed.

---

David Nathaniel P. Riego · BS Data Science, UA&P (Aug 2023 – Aug 2027 expected) · LinkedIn