AI-assisted market intelligence
Cote Assetage combines predictive modelling with a daily reporting cycle, so you can see how each signal performed rather than take a claim on trust.
The engine
The platform's edge is not a single clever prediction but a continuous process: data comes in, models produce probability-weighted views, and those views are checked against exposure and volatility before anything reaches you.
The engine reads order flow, volatility patterns and macro releases together, producing a probability-weighted view of likely price movement rather than a single directional call. Each signal carries a confidence range, so position sizing can reflect how certain the model actually is.
Exposure across open positions is recalculated continuously, not only at the point a trade is placed. If correlation between assets shifts, or volatility moves outside the range a signal was generated in, the position is flagged for review before it becomes a larger problem.
Once a trading session closes, model outputs are compared against what actually happened. Where a signal missed, the discrepancy feeds into the next training cycle, so the model is being checked against reality on an ongoing basis rather than left to run unexamined.
Underneath, the platform processes structured market data alongside unstructured inputs such as news sentiment and regulatory filings, using a mix of statistical and machine-learning models chosen for the type of question being asked. No single model is treated as authoritative: outputs are cross-checked against simpler benchmarks before they reach a report.
Methodology & reporting
Every trading day, the models generate a set of signals and risk assessments, which are logged before the session opens and left untouched until it closes. Nothing is edited or re-labelled afterwards, so the report you read the next morning reflects exactly what the model committed to, not a version smoothed over with hindsight.
Each report sets out which signals were generated, how confident the model was at the time, and what actually happened in the market. Where a signal underperformed, the report says so plainly, along with a brief note on the likely cause, whether that is a data gap, a regime shift, or a model limitation.
How each probability-weighted signal compared to what the market actually did.
Performance measured against a comparable passive benchmark, not in isolation.
The certainty range attached to each signal when it was generated, not adjusted afterwards.
A record of maximum drawdown and position exposure across the reporting period.
Which datasets, and which version of them, informed a given day's output.
Underlying datasets are versioned and timestamped, so any report can be traced back to the exact data snapshot that produced it. We do not retrospectively adjust historical outputs to make earlier performance look better.
From data to decision
Market data, order book depth, and relevant external sources are pulled in continuously and checked for gaps or anomalies before anything reaches a model.
Models produce probability-weighted signals with an associated confidence range, rather than a single fixed prediction.
Each signal is weighed against current exposure, volatility and correlation with existing positions, adjusting recommended sizing accordingly.
The output is a ranked recommendation with its reasoning attached, ready for a trader or investment committee to act on, adjust, or decline.
The framework is deliberately conservative about certainty. Where confidence is low or data is incomplete, the system says so rather than forcing a recommendation, leaving the final judgement with the person accountable for the outcome.
Applications
A day trader working several instruments at once uses the platform to flag rising correlation risk before it shows up in a drawdown. Signals update through the session, and each one is timestamped against the report published that evening, so performance can be checked trade by trade rather than taken on trust.
An investment committee reviewing quarterly capital allocation uses longer-horizon scenario modelling to stress-test proposed positions against a range of macro conditions. Because the reasoning behind each recommendation is documented, it can be presented to stakeholders alongside the numbers, not instead of them.
About Cote Assetage
Cote Assetage was built by a small team with backgrounds in quantitative research and market data engineering, working on the premise that most trading tools ask for a great deal of trust upfront and offer very little verification in return. Our starting point was the opposite: publish the reasoning, publish the confidence level, and publish the outcome, every day, so the model's track record speaks for itself.
The platform is designed to sit alongside your existing process rather than override it. Signals and reports are inputs to a decision, not a substitute for the judgement of the person making it.
Questions we're asked often
Models are trained on historical data, which is never entirely neutral. We address this by testing outputs against multiple time periods and market regimes rather than a single dataset, and by publishing confidence ranges so a low-certainty signal is never presented with the same weight as a high-certainty one.
Latency varies with the type of data involved: structured price and order-book data is processed on an ongoing basis through the session, while inputs such as filings or news sentiment take slightly longer to parse. We would rather be accurate about this variation than promise a single figure that does not hold for every data type.
No. Every signal in the daily report is accompanied by the reasoning behind it and the confidence level at the time it was generated. The intention is for the platform to extend your own analysis, giving you more to work with, not to replace your judgement with an opaque output.
Data is encrypted in transit and at rest, and access to raw datasets is logged and restricted to what a given process actually needs. We do not sell or share client data with third parties.
Yes. The live data demo shows a sample of historical signals alongside the outcomes recorded in that day's report, so the model's track record can be assessed before deciding whether it suits your approach.
Beyond encryption and access controls, every dataset used in a report is versioned, so if a question is ever raised about a specific day's output, we can show precisely which data produced it.
If you would rather see a model's actual track record than take a claim on faith, the live data demo shows a sample of past signals alongside the outcomes recorded in that day's report. There is no obligation attached to looking.