DCC: Differentiable Cardinality Constraints for Partial Index Tracking
Authors: Wooyeon Jo, Hyunsouk Cho
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we validate the proposed DCCfpp with various dataset in three aspects: 1) We compare index tracking errors to assess the performance of partial replication. 2) We measure the performance of the generated portfolio using commonly used metrics, 3) we compare the runtime of methods to highlight their efficiency. We provide implementation and data loading scripts1. 5.1 Experimental Settings Data We conduct experiments using the following three market indices: 1. S&P 100 Index 2. S&P 500 Index 3. KOSPI 100 Index |
| Researcher Affiliation | Collaboration | Wooyeon Jo1,3, Hyunsouk Cho1,2* 1Department of Artificial Intelligence, Ajou University, Suwon, Republic of Korea 2Department of Software and Computer Engineering, Ajou University, Suwon, Republic of Korea 3The AI Lab Inc., Seoul, Republic of Korea EMAIL |
| Pseudocode | No | The paper describes methods using mathematical formulations and descriptive text, but it does not contain any clearly labeled pseudocode blocks or algorithms. |
| Open Source Code | Yes | We provide implementation and data loading scripts1. 1Code and proof details: https://github.com/qt5828 |
| Open Datasets | Yes | We conduct experiments using the following three market indices: 1. S&P 100 Index 2. S&P 500 Index 3. KOSPI 100 Index For our analysis, we utilized data spanning from January 1, 2018, to April 30, 2023. The data was sourced from Yahoo Finance (Aroussi 2019). |
| Dataset Splits | Yes | The backtesting is conducted using a sliding window technique, where the data period is shifted by one day at a time. On rebalancing days, which are specific days set at regular intervals, the portfolio is adjusted by recalculating and applying new asset weights based on the most recent data. On other days, the performance is assessed using the weights fitted on the most recent rebalancing day. In our study, we rebalance the portfolio on a quarterly basis. For each rebalancing, we utilze one year of historical data to obtain the portfolio weights. |
| Hardware Specification | No | The paper mentions general concepts like |
| Software Dependencies | No | These mathematical optimization techniques are easily implemented using libraries such as CVXPY (Diamond and Boyd 2016) or Sci Py (Virtanen et al. 2020), which efficiently find precise solutions. |
| Experiment Setup | Yes | In our study, we rebalance the portfolio on a quarterly basis. For each rebalancing, we utilze one year of historical data to obtain the portfolio weights. The cardinality constraint was set to K = 25 and K = 40, respectively, and the tracking results are shown in Figure 4. As shown in Figure 6, when a ≤ 14, none of the conditions are met. For a ≥ 14, C0 is satisfied, C2 is satisfied for a ≥ 70, and finally, C1 is met when a ≥ 138, 157. a should be set to at least 138,157 to satisfy all conditions in Python s 64-bit floating-point precision, ensuring that our DCCfpp an accurately calculates the portfolio s cardinality and guarantees the enforcement of the cardinality constraint, regardless of the dataset. |