Optimasi Portofolio Black-Litterman dengan Rebalancing Berbasis Views dari Prediksi Return melalui Temporal Fusion Transformer

Putri, Sabrina Astriani (2026) Optimasi Portofolio Black-Litterman dengan Rebalancing Berbasis Views dari Prediksi Return melalui Temporal Fusion Transformer. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221034-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Portofolio investasi saham merupakan kombinasi beberapa aset yang disusun untuk mencapai keseimbangan optimal antar risiko dan return. Model klasik Markowitz yang umum digunakan memiliki kelemahan karena sangat sensitif terhadap kesalahan estimasi parameter input sehingga menghasilkan bobot yang tidak realistis. Penelitian ini mengembangkan metode optimasi portofolio dengan memanfaatkan views investor beserta tingkat ketidakpastiannya yang diperoleh dari prediksi return saham. Prediksi diperoleh menggunakan model Temporal Fusion Transformer (TFT) yang dilatih pada data historis sembilan saham periode 2020–2025 untuk menghasilkan prediksi berbasis kuantil. Nilai median (Q50) digunakan sebagai views, sedangkan rentang Q10–Q90 merepresentasikan ketidakpastian. Keduanya diintegrasikan ke dalam kerangka Black–Litterman bersama implied equilibrium return (risk aversion 2,5; τ = 1) untuk menghasilkan estimasi return posterior yang lebih stabil. Optimasi dilakukan dengan pendekatan mean–variance menggunakan Sequential Least Squares Programming (SLSQP) dengan kendala long-only dan fully invested. Kinerja dievaluasi melalui walk-forward backtesting berbasis rolling window kuartalan pada tahun 2025 menggunakan Sharpe ratio dan turnover. Model TFT menghasilkan prediksi presisi dengan rata-rata MAPE 2,08% dan coverage Q10–Q90 sebesar 79,66%. Hasil backtesting menunjukkan portofolio BL–TFT secara konsisten mengungguli portofolio pasar pasif di setiap kuartal, dengan akumulasi return 7,13% dibandingkan -0,44% pada pasar. Portofolio bersifat agresif mengejar imbal hasil di awal tahun (return 5,27%; Sharpe 0,2607 pada kuartal pertama) dan beralih ke mode perlindungan modal pada paruh kedua dengan menekan volatilitas hingga sekitar 7,7%, disertai turnover tinggi (mencapai 188,2%). Penelitian ini berkontribusi pada pengembangan optimasi portofolio di pasar modal Indonesia
=======================================================================================================================================
Stock investment portfolios combine multiple assets to achieve an optimal balance between risk and return. The classical Markowitz model, though widely used, is highly sensitive to estimation errors in its input parameters, often producing unrealistic weights. This study develops a portfolio optimization method that incorporates investor views together with their uncertainty, derived from stock return predictions. Predictions are generated using a Temporal Fusion Transformer (TFT) model trained on historical data of nine stocks over 2020–2025 to produce quantile-based forecasts. The median value (Q50) serves as the investor views, while the Q10–Q90 range represents uncertainty. Both are integrated into the Black–Litterman framework together with the implied equilibrium return (risk aversion 2.5; τ = 1) to produce more stable posterior return estimates. Optimization is carried out using a mean–variance approach with the Sequential Least Squares Programming(SLSQP) method under long-only and fully invested constraints. Performance is evaluated through walk-forward backtesting based on a quarterly rolling window during 2025, using the Sharpe ratio and turnover. The TFT model produced precise forecasts, with an average MAPE of 2.08% and a Q10–Q90 coverage of 79.66%. Backtesting shows that the BL–TFT portfolio consistently outperformed the passive market portfolio in every quarter, achieving a cumulative return of 7.13% versus -0.44% for the market. The portfolio is active and tactical: aggressively pursuing returns early in the year (5.27% return; 0.2607 Sharpe ratio in the first quarter) and shifting to capital preservation in the second half by suppressing volatility to around 7.7%, accompanied by high turnover (reaching 188.2%). This study contributes to the development of portfolio optimization in the Indonesian capital market

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Black-Litterman, IDX30, Optimasi Portofolio, Prediksi Return, Temporal Fusion Transformer, Portfolio Optimization, Return Prediction
Subjects:	H Social Sciences > HG Finance > HG4012 Mathematical models H Social Sciences > HG Finance > HG4529.5 Portfolio management H Social Sciences > HG Finance > HG4915 Stocks--Prices Q Science > Q Science (General) > Q180.55.M38 Mathematical models Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Sabrina Astriani Putri
Date Deposited:	17 Jul 2026 06:34
Last Modified:	17 Jul 2026 06:34
URI:	http://repository.its.ac.id/id/eprint/135319

Actions (login required)

View Item