📘 Using the forecasts wisely

Our best predictors get the direction of the next hour right about 53 times out of 100. That is a real statistical edge, but a thin one. Here is what the figure means, and what it does not mean. These are general principles for information, not advice tailored to your situation.

1. What "53%" means

• Over 100 hours, the forecast gets the direction (up or down) right about 53 times and wrong about 47 times. A coin flip would give 50.
• It is not "a 53% chance of making money" on a trade: moves are not all the same size, and fees apply to every purchase and every sale.
• It is an estimate, not a constant. For gold, 53.3% measured over 1,672 hours means the true value is probably between 50.9% and 55.7%. Unlike a casino's edge, which is known and permanent, this edge can shrink or disappear when markets change. That is why we publish the live accuracy, updated every hour.

2. Losing streaks are normal

With 47% wrong calls, runs of bad forecasts will certainly happen. Over 1,000 forecasts, the longest run of consecutive misses is typically 8 or 9.

Share of capital at risk per position	After 10 losses in a row
1%	90% left (−10%)
2%	82% left (−18%)
10%	35% left (−65%)

A 65% loss then needs a 186% gain just to break even. Professionals therefore generally limit the risk of each position to a small fraction of their capital.

3. The Kelly criterion, and its limits

The Kelly criterion computes the share of capital that maximizes long-term growth. With wins and losses of the same size, it equals win rate − loss rate: 53% − 47% = 6%. Many professionals use only a quarter of it (1.5%), because:

• the true win rate is uncertain: at 51%, Kelly drops to 2%, and at 50% it is zero;
• fees reduce the real edge;
• betting more than Kelly lowers growth and sharply raises the risk of ruin.

These figures illustrate a principle; they are not a sizing recommendation.

4. It takes many forecasts

Over 10 or 100 forecasts, luck dominates: a 3-point edge only shows after hundreds or even thousands of hours. A good or a bad week proves nothing. That is why we show the number of hours next to every percentage.

5. Costs can erase the edge

Friction	Effect	General good practice
Commissions and bid-ask spread	taken from every round trip	compare your platform's real fees
Slippage	filled at a worse price than shown	be wary of the minutes after major economic announcements
Delay	the forecast is for the current hour: acting late reduces its value	check the forecast time shown in the report

Over one hour, a market's average move is often small: round-trip fees of the same size can cancel an edge of a few points.

6. What we have checked, and what remains to be proven

• Walk-forward test on about 6 months of hourly data, April to October 2026. That is less than the 5 to 10 years, including crises, one would ideally want: the test will be extended as data accumulates.
• Paper trading (virtual money, real prices, fees included) started on October 7, 2026 for Bitcoin. We will publish its results.
• Live track record of every published forecast, on each predictor's page.

The usual caution before committing money: observe first, then, if you do, start very small and only scale up if real results confirm the statistics.