Walk-forward study

How the models actually performed

Every number here comes from walk-forward testing on historical data: the model is trained on the past and scored on data it has never seen. Some of it is not what we hoped for. We publish all of it, because a research tool that only shows you its wins is a brochure.
Twelve markets  ·  three horizons  ·  59,000+ scored observations  ·  out of sample
Read this as a study, not a scoreboard. These are historical results, not a record of live trading, and we keep the two apart on purpose. Test windows overlap, so the effective sample is much smaller than the row count: treat a gap of one or two points as noise rather than as a ranking. Past testing does not guarantee future results.
The range is the result, the price target is not
What held up
The strongest and most reliable finding in the whole study

When we published a 90% price range, price landed inside it close to 90% of the time, across every market and horizon we serve. Measured coverage ran from 87.9% to 91.0% against a nominal 90%. That is exactly what a calibrated range is supposed to do, and it is the claim we stand behind.

The opposite result matters just as much. Our forecast of the specific centre price never beat a simple random walk, in any of the twelve markets. Predicting roughly where price will settle added nothing over assuming it stays put. So we stopped selling it. You will never see a price target on Enodara, because we could not earn one.

Directional accuracy, every market and horizon
The full table, including what failed
Our publication floor is 55%. Green clears it, amber is marginal, grey falls short. Where a market and a horizon do not clear, the model stands aside and publishes a range instead of a direction.
MarketClass 5-minute Hourly Daily Natural clock
AMDStock57.853.1no edge5-minute
NVIDIAStock57.050.8no edge5-minute
AppleStock55.754.0no edge5-minute
BitcoinCrypto56.255.9no edge5-minute and hourly
EthereumCrypto55.757.3no edgehourly, strongest overall
SilverMetal55.255.1no edgeflat across the board
GoldMetal55.055.7no edgehourly
Oil (WTI)Energy54.256.1no edgehourly
BrentEnergy53.156.2no edgehourly
Dow JonesIndex54.055.8no edgehourly
Nasdaq 100Index54.455.6no edgehourly
S&P 500Index52.654.2no edgeweak everywhere

Each asset class turns out to have a natural clock. Single stocks are strongest over five minutes and close to a coin flip over an hour, because their edge is short-horizon and washes out. Energy and indices are the mirror image. Nothing in the study showed reliable directional skill on the daily chart, so the daily view publishes a range only.

What a conviction figure means
Calibrated, which means lower than you expect
This page previously said something different, and it was wrong

Conviction is calibrated, so the number is meant to be read as a real probability rather than as enthusiasm. In practice that puts almost every reading much closer to 50% than people expect. Readings sit on roughly a 45 to 50, 50 to 55, and 55%-plus scale, and even the strongest bucket lands near 58%.

An earlier version of this page reported accuracies in the eighties for high-conviction readings. Those came from raw, uncalibrated model output measured on a small number of instances, and the engine that produced them has since been recalibrated. We do not publish those figures any more because the current system cannot produce them, and a number you can never see in the product has no business on a page about performance.

What did not work
The findings we would rather not have
Published because leaving them out is what makes the rest untrustworthy
What this study covers

This study measured the twelve markets we covered when it was run. Our coverage has since grown to thirty-four, and those newer markets are not in the table above. We would rather leave a gap than fill it with numbers this study did not measure, so they will appear once they have been through the same process. The figures here will be refreshed when the models are next retrained and rescored.

Past model performance does not guarantee future results. Markets are non-stationary, so model accuracy may vary across different market regimes. Enodara provides market intelligence for informational and educational purposes only and does not constitute financial advice. Enodara is not a registered broker-dealer, investment adviser, or financial advisor, and does not manage client funds or execute trades. All figures will be updated following each model retrain.