No, not at all, given the context of stock trading. Stocks do not trade the same way today as they did in 2014. Similarly there is no point in using trading data from the 1850s given that that kind of market with those kinds of traders will never ever exist again. You can only pick a few recent months or weeks to capture current trading sentiment/technique and even then everything could get blown away after the next rate hike or international incident.

Generally I don't think there is any alpha in training transformers to predict the next price point just given historical price data, because the price is determined by humans (and algorithms trained on data generated by humans) that react to news. If you can predict the news, you can probably predict stock prices, but if you could predict the future you'd have AGI and not some dingy time series calculator.

For rough, high-level comparisons, it might be seen as "not far off," but for detailed, technical assessments, the difference is considerable.[1]

[1] https://chat.openai.com/share/a19a3b57-398c-49e7-a140-f58784...

Good enough. The comparison is silly though: time series data is not anything like tokenized text data.