DatedGPT: Preventing Lookahead Bias in Large Language Models with Time-Aware Pretraining
This addresses a critical problem in financial forecasting for practitioners by preventing data leakage, though it is an incremental adaptation of existing methods with temporal constraints.
The paper tackles lookahead bias in financial backtesting by introducing DatedGPT, a family of twelve 1.3B-parameter language models trained on temporally partitioned data with annual cutoffs from 2013 to 2024, which effectively bounds knowledge to each cutoff year while maintaining competitive performance on benchmarks.
In financial backtesting, large language models pretrained on internet-scale data risk introducing lookahead bias that undermines their forecasting validity, as they may have already seen the true outcome during training. To address this, we present DatedGPT, a family of twelve 1.3B-parameter language models, each trained from scratch on approximately 100 billion tokens of temporally partitioned data with strict annual cutoffs spanning 2013 to 2024. We further enhance each model with instruction fine-tuning on both general-domain and finance-specific datasets curated to respect the same temporal boundaries. Perplexity-based probing confirms that each model's knowledge is effectively bounded by its data cutoff year, while evaluation on standard benchmarks shows competitive performance with existing models of similar scale. We provide an interactive web demo that allows users to query and compare responses from models across different cutoff years.