After 112 resolved public forecasts, one result is already clear:
High-conviction signals are working. Low-conviction signals are not.
It is still early. This is not enough data to make sweeping claims. But it is enough to test whether TradeHorde's conviction layer is doing real work.
So far, it is.
How TradeHorde works
TradeHorde uses a two-layer forecasting process.
Multiple models analyze the same ticker independently, arguing both the bull and bear case. An arbiter layer then decides whether the best call is long,
short, or pass.
When it takes a side, it also assigns a conviction score.
That score is supposed to separate stronger forecasts from weaker ones. If it were just decoration, performance across conviction tiers should blur together.
They do not.
The first result that matters
Across 112 resolved forecasts, the conviction tiers behave exactly the way you would hope.
| Tier | N | Win Rate | Avg R-Multiple | Profit Factor |
|---|---|---|---|---|
| High Conviction | 20 | 65.0% | +0.81R | 6.51 |
| Conditional | 34 | 61.8% | +0.98R | 2.81 |
| Low | 58 | 36.2% | +0.20R | 0.85 |
The most important number there is profit factor.
Below 1.0, a strategy loses money in aggregate. Low-conviction signals are below that line at 0.85. High-conviction signals are far above it at
6.51.
That is what a useful filter looks like.
The combined equity curve tells the story
Compounding matters more than any one metric in isolation.
When the resolved forecasts are grouped into conviction-tier portfolios and plotted on one equity curve, the picture becomes obvious:
| Filter | Compounded Return | Expectancy |
|---|---|---|
| All Tiers | +208.8% | +1.18% |
| High Only | +92.8% | +3.49% |
| Conditional Only | +117.4% | +2.46% |
| High + Conditional | +319.3% | +2.84% |
| Low Only | -26.4% | -0.37% |
The combined equity curve shows the main result clearly: low-conviction signals drag performance, while High + Conditional outperforms the all-tier baseline.
The implication is simple:
The edge may not come from trading more. It may come from refusing the weakest forecasts.
Why that matters
This is the encouraging part of the early dataset.
TradeHorde does not just produce forecasts. It appears to rank them in a way that matters.
As conviction drops, performance deteriorates:
- High performs best
- Conditional remains strong
- Low becomes a drag
That suggests the scoring layer is useful not just for deciding what to trade, but what to ignore.
Calibration is early, but promising
There is also an early calibration signal.
Across the first 48 scored forecasts since the scoring window opened, the raw Brier score is 0.244 versus 0.250 for random guessing.
That edge is still small and needs more data. But directionally, it is encouraging.
The conviction score is not behaving like pure noise.
What comes next
The next phase is live validation.
TradeHorde's Portfolio Mode is now tracking multiple strategies in real time, including:
- all-tier baseline portfolios
- conviction-filtered portfolios
- an AI directional gate that makes a live yes/no decision before entry
SPY options execution is also live, with regime-aware spread structures and active risk management.
The first 112 resolved forecasts do not settle the question.
But they do establish something important: TradeHorde's conviction layer appears to matter, and filtering weak signals may be one of the clearest edges in the system so far.
All equity curves shown are equal-weight hypothetical returns using 1 unit per signal. Past performance does not guarantee future results. TradeHorde is in active development.
TradeHorde is building a public market forecasting engine that keeps score. Follow along at @tradehorde and track the next 100 forecasts as live results come in.