Skip to content

Process

From hypothesis to live validation

A model begins with a documented research question. Research and backtesting lead to live paper trading, then real-money live trading, our current stage.

Research papers Enquiries

The research process

Four stages, in order. The internal paper records the evidence, limitations and research conclusions.

Research question

QUESTION, HYPOTHESIS · EVIDENCE AND METHODS

QUESTION, HYPOTHESIS · EVIDENCE AND METHODS

A model starts as a written research question with an economic hypothesis, set against prior research. The paper records where each piece of data comes from and when it became available, how the sample is defined, the methods and the testing protocol.

Validation

UNSEEN PERIODS · RESULTS AFTER COSTS

UNSEEN PERIODS · RESULTS AFTER COSTS

Validation asks whether the result holds on periods that were not used to build it, with trading, financing and short-borrow costs included where relevant. Robustness checks and competing explanations belong in the record.

Model portfolio analysis

RISK AND OVERLAP · PORTFOLIO CONTRIBUTION

RISK AND OVERLAP · PORTFOLIO CONTRIBUTION

A model that looks good alone may add little to a hypothetical combined portfolio. Correlation, overlapping positions and common-factor exposure are weighed against its marginal contribution to the whole.

Research output

DOCUMENTED MODELS · LIMITATIONS AND REVIEW

DOCUMENTED MODELS · LIMITATIONS AND REVIEW

The output is a documented model, with assumptions, limitations and criteria for further review. Any future product implementation would depend on separate agreements, responsibilities and applicable regulatory requirements.

From research to real-world evidence

The development path moves through four distinct stages. We are now in real-money live trading.

  1. Research

    Economic hypothesis

    Define the economic idea, data and rules, and document what could make the model fail.

  2. Backtest

    Historical simulation

    Test on unseen historical periods, with explicit trading, financing and borrow-cost assumptions.

  3. Live paper trading

    Real-time simulation

    Run the process on current market data with simulated capital to test operations and execution assumptions.

  4. Live real-money trading

    Current stage

    Trade with real capital to assess actual execution, costs and model behaviour. Capital is at risk.

We plan to build at least 12 months of real-money live trading evidence before a readiness decision on any investor launch. Reaching 12 months does not trigger a launch: the evidence, operating arrangements and applicable requirements must also support it.

Track-record verification

We use AuditZK for track-record verification. Verification does not guarantee future performance.

Hypothetical
Hypothetical, illustrative extract. Full papers and model definitions remain internal.

Inside a research paper

Question and hypothesis

Economic rationale and prior research.

What is the question, why should the effect exist, and what has earlier research found?

A complete paper opens with an abstract, then states the research question, the economic hypothesis that would explain the effect and the related literature.

Data and methods

Availability dates, sample and test protocol.

Which data, as known on which date; which companies and periods; how the test is run.

A complete paper contains the experimental design: the data and its timing, how the sample is defined, the methods and the testing protocol.

Evidence and challenge

Out-of-sample results, costs and robustness.

Does the result hold on unseen data, after costs, and against competing explanations?

A complete paper reports the results, then the falsification and robustness tests: deliberate attempts to show that the result is fragile, wrong or better explained by something else.

Conclusions and review

Limitations, failure conditions and review.

What would show that the hypothesis no longer holds, and how the model is reviewed.

A complete paper sets out its limitations, research conclusions and criteria for further testing.

Appendix

Sources, timestamps and reproducibility.

A complete paper records a source and timestamp audit, showing where each input came from and when it became available, and ends with a reproducibility appendix.

Validation through time Learn from the past. Test on the next period.

The standard for allocation models is walk-forward evaluation: each model is fitted only on information available before its test period, then evaluated on the period that follows; the window then expands and the test repeats. Comparisons must apply trading, financing and borrow costs consistently, and use simple baselines such as risk parity and static weights.

Past-only inputs Returns after full costs Risk-parity comparison
Market regimes
Hypothetical
Hypothetical walk-forward schematic. Blocks show sequence, not dates, test results or an allocation advantage.
Hypothetical
Hypothetical illustration of changing market conditions. It contains no data, regime probabilities or strategy weights.
Model research

Market regimes and model weights

Our research explores machine learning to estimate recurring market regimes. Conditional estimates of return, risk and correlation can inform hypothetical model weights, which are tested within specified risk constraints.

Historical data

Simulated model returns and historical risk features, using only information that was available at the time.

Regime learning

Recurring market conditions, estimated from the history of those risk features.

Model weights

Hypothetical weights within risk constraints. Estimates for each condition can inform model weights; simulated exposure, concentration, trading liquidity, capacity and turnover constrain them.

Research review

Costs, stability and review. A regime model must be judged on periods that come after the ones it learned from.

RISK FEATURES

Volatility Macro liquidity Credit Rates Breadth Cross-asset conditions

This describes research into an allocation method. No improvement in results is claimed, and no measured regime probabilities, live allocations or model weights are shown.

Complementary data for research and validation

Three families of data support research, validation and implementation modelling, complemented by macroeconomic series and regulatory bank reports.

Exchange data up to Level 3

Trades, quotes and order-by-order books, for liquidity analysis and execution modelling.

Level 3 availability depends on the venue and dataset.

Point-in-time U.S. financials

As-reported statements aligned to filing dates: point-in-time data for fundamental and accounting-quality research.

Global coverage and earnings expectations

Analyst estimates, actuals and earnings surprises, with global prices and company coverage.

Data providers

When it was known matters as much as what was known

A result built on information nobody had at the time is not evidence. Research must therefore use data as it was available on each date: financial statements from their filing dates, and other variables lagged to their release.

Validation through time
Hypothetical
Hypothetical schematic. No company, figure or date is represented.

What hedging leaves in place

Next: risk

How risk is considered across the whole portfolio, and the risks no framework removes.

Risk Common questions