Introducing T+5 Forecasts: Five-Day Horizon Predictions for S&P 500 Stocks

Why a Five-Day Horizon?

Most retail investors don't trade every single day. They look at their portfolio a few times a week, decide whether to add to a position or trim it, and then wait. A one-day forecast is useful for active traders. A five-day forecast is useful for everyone else.

T+5 means the forecast predicts where a stock's price will be at close of the fifth trading day from now — roughly one calendar week out. This is a fundamentally different prediction problem from T+1. The signal-to-noise ratio drops as the horizon extends, but the practical utility increases for the majority of our users.

The Data Challenge: Error Compounds Over Time

The core challenge of multi-day forecasting is error compounding. If your T+1 model makes a small directional error, that error propagates into your T+2 estimate, which compounds into T+3, and so on. By T+5, a naive autoregressive approach has accumulated so much error that the forecast band becomes essentially uninformative.

There are two ways to address this. The first is to train a direct multi-step model: instead of predicting one day at a time and chaining the predictions, you train a model to directly predict the T+5 close from today's features, bypassing the chain altogether. The second is to use a sequence-to-sequence architecture that simultaneously predicts all five days, allowing the model to learn how multi-day trajectories typically develop without autoregressive error compounding.

We use both approaches and ensemble the results, consistent with our overall dual-model philosophy. Our direct LSTM predicts T+5 directly from the current feature window. Our seq2seq model predicts the full five-day trajectory. The ensemble uses the same dynamic weighting approach as our T+1 system.

Training Labels for Multi-Day Forecasting

For T+1 models, the training label is simply: did the next close go up or down? For T+5, the label is the cumulative return over the five-day window. But this creates a problem: a stock can go up 3% over five days while spending days 2 and 3 down 2% each. The five-day label says "up" but the path was volatile.

We found that training on raw five-day returns produced models that were good at predicting end-of-week closes but poor at predicting whether the stock would be volatile on the way there. For the forecast band to be useful, we also need to capture intra-week volatility.

Our solution was to train a secondary volatility model specifically for the T+5 horizon. This model predicts the expected range of daily returns over the five-day window, not just the end point. The T+5 forecast band is widened by the predicted intra-week volatility, giving users a more accurate sense of how much the stock might move in either direction before reaching the five-day target.

Accuracy on a Five-Day Horizon

The honest answer is that five-day accuracy is lower than one-day accuracy, and it always will be. Markets are efficient over longer horizons in a way they are not over very short horizons. The edge available to a predictive model over one trading day is real but small. Over five days, macro events, news flow, and earnings surprises can overwhelm any technical signal entirely.

Our T+5 Met/Beat rate across the S&P 500 is lower than our T+1 rate. We are transparent about this on the platform — each forecast card shows separate one-day and five-day accuracy statistics for the stock, so you can see exactly how the model performs at each horizon on that specific name.

For a subset of stocks — primarily large-cap, high-liquidity names with stable institutional ownership and low exposure to idiosyncratic news risk — the T+5 model performs surprisingly well. These tend to be stocks in defensive sectors where weekly price moves are driven more by systematic factors (interest rates, sector rotation) that have more predictable dynamics than individual company news.

How to Use T+5 Forecasts in Practice

We recommend using T+5 forecasts primarily for position sizing and entry timing decisions, not as standalone trading signals. If the T+1 and T+5 forecasts both point in the same direction with high confidence, that alignment is a stronger signal than either alone. If they disagree — the T+1 forecast is bearish but the T+5 forecast is bullish — the stock may be expected to dip short-term before recovering.

The most practical use case is for deciding whether to enter a position now or wait. If you want to buy a stock but the T+1 forecast shows a high-confidence downward move, the T+5 view can tell you whether the model expects that dip to reverse within the week, suggesting a better entry point may exist in the near term.

All T+5 forecasts are visible on every stock detail page alongside the T+1 forecast. The five-day forecast chart shows the predicted direction, the band, and the confidence score for the weekly horizon. You can toggle between T+1 and T+5 views to compare the model's short-term and medium-term outlook on any S&P 500 stock.