How We Score Every Forecast We Make

Why Public Scoring Matters

Every AI stock forecasting tool claims high accuracy. Few publish the underlying data. At OptiHedge, we believe the only way to earn investor trust is to publish every single forecast verdict — wins and misses alike — in a permanent, searchable scorecard.

When you open any stock on OptiHedge, you see two things immediately: the model's current prediction and a running score of how that model has performed historically on that same stock. You don't have to take our word for accuracy — you can verify it yourself before acting on any signal.

The Three Verdicts: Bullseye, Outperform, Miss

At the close of each trading day, our system checks the actual closing price against the forecast made the prior evening. Each forecast is scored as one of three outcomes:

• Bullseye — The actual close landed inside the predicted price band. The model not only called the direction correctly but also estimated the magnitude accurately enough to capture the move within its confidence interval.
• Outperform — The close moved beyond the upper band in the direction the model predicted. The direction was right, the magnitude was underestimated. Still a useful signal.
• Miss — The actual close moved opposite to the forecast, or landed outside the band in the wrong direction. The model called it wrong.

What Is the Met/Beat Rate?

The Met/Beat rate is our primary accuracy metric. It counts the percentage of forecasts that received either a Bullseye or Outperform verdict — in other words, forecasts where the actual price moved in the predicted direction, regardless of whether it stayed inside or went beyond the band.

A Met/Beat rate of 65% on AAPL means that over the historical period shown, 65 out of 100 daily forecasts for Apple called the direction correctly. It is a directional accuracy metric, not a magnitude metric.

The Role of the Confidence Score

Not all forecasts are equally certain. Our ensemble model outputs a confidence score from 0 to 100 alongside each prediction. High-confidence forecasts (above 70) have historically shown higher Met/Beat rates across the S&P 500. Low-confidence forecasts (below 40) are still displayed but carry a visible warning that the model's signal is weaker than average.

We recommend filtering your watchlist to stocks where the model confidence is above 60 before acting on any single forecast. The platform lets you sort by confidence directly from the dashboard.

Volatility-Adjusted Bands

The predicted price band is not a fixed-width range. It is dynamically calculated based on the stock's recent volatility. A low-volatility stock like JNJ will have a tighter band than a high-volatility stock like NVDA on any given day. This means a Bullseye on NVDA requires a more accurate magnitude prediction than a Bullseye on JNJ.

We apply a minimum band width floor equal to one standard deviation of the stock's 20-day rolling return. This prevents artificially high Bullseye rates on extremely low-volatility days where almost any close would land inside a very narrow band.

The 90-Day Scorecard

Every stock page shows a 90-day rolling scorecard broken down by verdict type. You can see the trend — whether accuracy has been improving or degrading over recent weeks. This is important because model performance is not static. Accuracy on a given stock can shift when its trading dynamics change (for example, after an earnings event or a major product announcement that changes how institutional investors position around it).

The 90-day window gives you enough data to assess whether the model has been consistently reliable on a stock recently, rather than relying on historical performance from a year ago.

Why We Don't Cherry-Pick

All forecasts are locked 15 minutes before market open. No forecast can be retroactively edited. The verdict is computed automatically by comparing the locked forecast to the official end-of-day closing price from the exchange. No human can intervene in the scoring process.

This matters because cherry-picking — publishing only the forecasts that turned out to be correct — is the most common form of manipulation in AI trading tools. By locking forecasts before open and scoring them automatically, we make cherry-picking structurally impossible.

What This Means for You as an Investor

Use the scorecard before adding any stock to your watchlist. If the 90-day Met/Beat rate on a stock is below 50%, the model has been essentially guessing on that stock recently — and you should weigh the current forecast accordingly. If the rate is above 65% with a recent upward trend, the model has been particularly well-calibrated on that stock.

No AI forecast is a guarantee. OptiHedge is a signal, not financial advice. But a transparent, auditable signal with a public track record is fundamentally more useful than a black box that only publishes its wins.