Optimization of Dynamic Crypto Portfolios via Deep Reinforcement Learning Under Transaction Costs
The DRL-PPO strategy generated an annualized return of 37.22% under a 0.1% transaction cost constraint.
Company · 6 documents
The DRL-PPO strategy generated an annualized return of 37.22% under a 0.1% transaction cost constraint.
BIS Working Paper No. 1049 assembled a Sensor Tower panel of retail crypto-exchange app downloads and daily active users (DAUs) covering 95 countries from August 2015 to June 2022.
In Vietnam, a one-standard-deviation increase in cryptocurrency downside is associated with a 2.3-basis-point lower VN-Index opening return and a 14.9-basis-point lower intraday return.
The dataset includes 1,095 daily observations per cryptocurrency, generating a total of 6,570 observations for analysis in EViews.
The proposed DAC-CPPO model achieves an overall accuracy of 97.6%, an MAE of 0.074, and an RMSE of 0.081 for portfolio optimization.
A systematic backtest on 26 usable post-listing restricted-share events from 2020 through 2025 evaluated short-selling returns around lockup expirations.