Peramalan Posisi Relatif Kinerja Return Menggunakan Improve Gated Recurrent Unit Berbasis Market Latent State dengan Multi-Head Cross-Attention

Eiffelin, Friska Naya (2026) Peramalan Posisi Relatif Kinerja Return Menggunakan Improve Gated Recurrent Unit Berbasis Market Latent State dengan Multi-Head Cross-Attention. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221016-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Saham merupakan salah satu instrumen investasi yang diminati masyarakat Indonesia karena menawarkan potensi return yang tinggi. Beberapa indeks sektoral BEI, seperti IDXFINANCE dan IDXENERGY menunjukkan kinerja yang lebih baik dibandingkan JCI dan LQ45 sehingga menarik bagi investor aktif dalam melakukan proses pemilihan saham (stock selection). Namun, peramalan return saham secara individual menjadi kurang efisien ketika jumlah saham yang dianalisis cukup banyak. Oleh karena itu, penelitian ini bertujuan untuk menganalisis karakteristik saham pada indeks IDXFINANCE dan IDXENERGY, mengevaluasi kemampuan metode MCI-GRU dalam memprediksi posisi relatif kinerja return saham, serta menghasilkan peramalan posisi relatif return saham untuk lima periode ke depan. Data yang digunakan berupa data harian harga pembukaan, penutupan, harga tertinggi, harga terendah, dan volume perdagangan periode 3 Januari 2022 hingga 31 Desember 2025. Metode MCI-GRU mengintegrasikan Improved GRU untuk menangkap pola temporal, Graph Attention Network (GAT) untuk memodelkan hubungan antar saham, serta multi-head cross-attention untuk merepresentasikan kondisi pasar laten. Hasil analisis menunjukkan bahwa IDXFINANCE memiliki pergerakan indeks yang lebih fluktuatif, sedangkan pada level saham individual IDXENERGY cenderung memiliki volatilitas yang lebih tinggi. Model terbaik pada IDXFINANCE diperoleh pada konfigurasi neuron GRU 32, hidden size GAT pertama 32, hidden size GAT kedua 4, attention head 6, dan latent state 32 dengan nilai Rank IC sebesar 0,0242. Sementara itu, model terbaik pada IDXENERGY diperoleh pada konfigurasi neuron GRU 64, hidden size GAT pertama 32, hidden size GAT kedua 4, attention head 6, dan latent state 32 dengan nilai Rank IC sebesar 0,0159. Nilai tersebut tergolong cukup baik dalam permasalahan pemeringkatan saham dan didukung oleh hasil backtesting yang menghasilkan Sharpe Ratio positif masing-masing sebesar 0,44 dan 0,42, serta nilai Maximum Drawdown yang masih terkendali sebesar 4% dan 11%. Selain itu, hasil peramalan lima periode ke depan menunjukkan pola pemeringkatan saham yang relatif konsisten pada kedua sektor. Hasil penelitian menunjukkan bahwa metode MCI-GRU mampu digunakan untuk meramalkan posisi relatif kinerja return saham dan dapat mendukung proses pengambilan keputusan investasi.
=====================================================================================================================================
Stocks are among the most attractive investment instruments in Indonesia due to their potential to generate substantial returns. Several sectoral indices on the Indonesia Stock Exchange, particularly IDXFINANCE and IDXENERGY have outperformed both the JCI and LQ45, making them appealing targets for active investors engaged in stock selection. However, forecasting individual stock returns becomes increasingly inefficient as the number of stocks under consideration grows. Therefore, this study aims to analyze the characteristics of stocks within the IDXFINANCE and IDXENERGY indices, evaluate the capability of the Multi-Head Cross-Attention Improved Gated Recurrent Unit (MCI-GRU) model in predicting the relative ranking of stock returns, and generate forecasts of relative stock return performance for the subsequent five trading periods. The dataset consists of daily open, high, low, close, and trading volume data spanning from January 3, 2022, to December 31, 2025. The proposed MCI-GRU framework integrates an Improved GRU to capture temporal dependencies, a Graph Attention Network (GAT) to model inter-stock relationships, and a multi-head cross-attention mechanism to learn latent market states. The results reveal that IDXFINANCE exhibits more pronounced fluctuations at the index level, whereas individual stocks within IDXENERGY demonstrate higher volatility. The best-performing model for IDXFINANCE was obtained using 32 GRU neurons, first- and second-layer GAT hidden sizes of 32 and 4, six attention heads, and 32 latent states, achieving a Rank Information Coefficient (Rank IC) of 0.0242. For IDXENERGY, the optimal configuration employed 64 GRU neurons with the same GAT and attention settings, yielding a Rank IC of 0.0159. These results indicate satisfactory ranking performance and are further supported by positive backtesting outcomes, with Sharpe Ratios of 0.44 and 0.42 and Maximum Drawdowns of 4% and 11% for IDXFINANCE and IDXENERGY, respectively. Furthermore, five-period-ahead forecasts exhibit relatively consistent stock-ranking patterns across both sectors. Overall, the findings demonstrate that MCI-GRU is effective for forecasting the relative performance of stock returns and can serve as a valuable tool to support investment decision-making.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	IDXENERGY, IDXFINANCE, MCI-GRU, Prediksi, Return, IDXENERGY, IDXFINANCE, MCI-GRU, Prediction, Return.
Subjects:	H Social Sciences > HA Statistics H Social Sciences > HA Statistics > HA30.3 Time-series analysis H Social Sciences > HG Finance H Social Sciences > HG Finance > HG4529 Investment analysis H Social Sciences > HG Finance > HG4915 Stocks--Prices Q Science > Q Science (General) > Q325.78 Back propagation
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Friska Naya Eiffelin
Date Deposited:	20 Jul 2026 02:25
Last Modified:	20 Jul 2026 02:25
URI:	http://repository.its.ac.id/id/eprint/135514

Actions (login required)

View Item