PRICE: A Systematic Study of LLM Adaptation Choices for Bitcoin Price Forecasting
The study introduces PRICE, a structured approach for adapting LLMs to short-term Bitcoin price forecasting, built on a 4-bit quantized LLaMA-3 8B model. It investigates how fine-tuning, numerical representation, prompting, inference, and decoding jointly influence forecasting performance. PRICE integrates Parameter-efficient fine-tuning with Low-Rank Adaptation (LoRA), Recursive multi-step inference, Integer-rounded numerical representation, Context-Task-Format (CTF) prompting, and Exact zero-temperature decoding. Ablation studies show that each component contributes to forecasting accuracy and reliability. LoRA enables efficient training on limited hardware, recursive inference improves accuracy, integer-rounded values reduce errors, CTF prompting outperforms Chain-of-Thought, Implicit Chain-of-Thought (iCoT), and few-shot prompting, and zero-temperature decoding improves stability.
PRICE is a systematic study of LLM adaptation choices for Bitcoin price forecasting, using a 4-bit quantized LLaMA-3 8B model. It combines LoRA fine-tuning, recursive multi-step inference, integer-rounded numerical representation, CTF prompting, and zero-temperature decoding. Ablation studies demonstrate that each component contributes to forecasting accuracy and reliability, with CTF prompting outperforming other prompting methods.
The combination of LoRA fine-tuning, recursive inference, integer-rounded numerical representation, CTF prompting, and zero-temperature decoding improves forecasting accuracy and reliability. CTF prompting outperforms Chain-of-Thought, Implicit Chain-of-Thought, and few-shot prompting. Integer-rounded values reduce errors, and zero-temperature decoding improves stability.
This research suggests that LLMs can be effectively adapted for financial time series forecasting with careful design choices, potentially enabling more accurate short-term cryptocurrency price predictions. The use of 4-bit quantization and LoRA indicates feasibility on limited hardware, which could lower barriers for financial institutions to deploy LLM-based forecasting models.
The approach could provide financial analysts and traders with improved short-term Bitcoin price forecasts, potentially enhancing trading strategies and risk management. The efficient training method may reduce computational costs for deploying LLM-based forecasting models.
Future work may explore applying PRICE to other financial assets or time series domains, and investigate further optimization of adaptation components. The findings could lead to more robust LLM-based forecasting systems in volatile markets.