Benchstreet is a curated collection of time series prediction models designed to help developers evaluate and compare the performance of different approaches in one-shot, long-term financial data forecasting.
The models are trained on 20 years of S&P 500 daily closing prices provided by Investing.com.
Important
This is not an objective benchmark! It's intended as a qualitative guide and a reference on how to implement these models.
| Model Type | |
|---|---|
| Transformer/Foundation Models | TimesFM (baseline • fine-tuned), Chronos (baseline • fine-tuned) |
| Feedforward Neural Networks (FNNs) | MLP (recursive • vector), N-BEATS (direct) |
| Convolutional Neural Networks (CNNs) | 1D-CNN (recursive • vector), TemporalCN (vector) |
| Recurrent Neural Networks (RNNs) | LSTM (recursive • vector • encoder-decoder), GRU (recursive • vector) |
| Statistical Models | ARIMA (recursive), SARIMAX (vector), FBProphet (direct) |
Want a model added to this list? Raise an issue here or make a PR!
Tip
The winner: N-BEATS. High accuracy with extremely low training time.
timesfm/fine_tune.py • download on huggingface 🤗
chronos/fine_tune.ipynb • download on huggingface 🤗
Want a model added to this list? Raise an issue here or make a PR!





































