Choosing portfolio risk software for a hedge fund

The requirement nobody writes on the RFP

Evaluating risk software requires more than a list of available statistics. Check how the system reacts to a changed portfolio, an unpriced position and a request to reconstruct an earlier report. Those cases reveal its assumptions and evidence boundaries.

Ask whether a reviewer can reconstruct a reported figure from the retained holdings, data dates, model settings and method. Include that exercise in an evaluation alongside the model features and export formats.

• The number changed and nobody can attribute the change to a cause.
• The report and the cockpit disagree, because they are two computations.
• The methodology is a PDF written three model versions ago.
• A position type the fund actually holds is silently excluded from the aggregate.
• Re-running last quarter's analysis is impossible because the data has moved underneath it.

The evaluation checklist

1. One computation, or several?

Ask whether the portfolio view, the deep-linked risk view, the PDF report and the API all call the same computation. Separate netting implementations can produce inconsistent results. Ask where sleeve netting happens and compare the output across those surfaces.

2. Is the arithmetic published?

Ask for the arithmetic and its capital basis. Gross exposure divided by net asset value answers a different question from gross exposure divided by invested capital. Cash and leverage affect that comparison. TradePolaris publishes the risk-limit calculations so a reviewer can inspect the convention.

3. What happens to instruments it does not price?

Test the treatment of an unpriced position during an evaluation. Compare the response with the documented coverage policy and inspect the exported report for the same labels.

Ask for a demonstration with an unsupported instrument and a name without enough history. Inspect what happens to the aggregate and which checks become unavailable. Position count and weight coverage describe different gaps. A report should disclose both its coverage basis and any reweighting of the modeled subset.

4. Ex-ante or ex-post?

Realized volatility summarizes historical portfolio returns. Ex-ante risk applies current weights to a covariance model, so it changes when the holdings change. For a fund that trades, check which basis each figure uses and how trading affects it.

5. Does it survive the size of a real book?

A sample covariance estimate can be singular when the history contains too few independent observations relative to the asset count. Factor models and covariance shrinkage are different ways to address estimation problems. Ask which method the system uses, its assumptions and when it changes methods.

6. Can the analysis be reopened?

Whether last quarter's number can be regenerated depends on whether the dataset version was pinned to the result. If the system reads “latest” and the source restated, the answer changes and nobody can explain why. See point-in-time data and survivorship bias for the underlying mechanics.

7. Can the team review the same analysis?

Ask the analyst, CIO, COO and reporting team to review the same portfolio calculation during the evaluation. Check whether they can inspect its inputs, method, coverage and export, and trace any changes between versions.

The measures a fund book actually needs

• Ex-ante VaR and CVaR, recomputed as the book changes. CVaR matters more than VaR for anything with option-like payoffs, because VaR is silent about the shape of the tail beyond it.
• Factor decomposition should distinguish common-factor exposure from specific risk. Review the covariance assumptions and residual coverage as well as the factor labels. Read stress testing and factor attribution for the reconciliation checks.
• Concentration, on both the obvious axis (largest positions) and the one that bites (largest RISK contributors, which are rarely the same names).
• Stress scenarios relevant to the fund's exposures, with historical and hypothetical assumptions stated. See stress testing and factor attribution.
• Liquidity, expressed as days to exit at a stated participation rate, per position and for the book.
• Sleeve-level netting, so a multi-strategy book reports what it actually holds after the sleeves offset each other rather than the sum of the parts.

What the allocator is actually checking

Include questions on independence, reproducibility and consistency in an operational review. Ask for evidence from a retained report and its inputs.

1. Independence. Is risk computed by something other than the person who put the trade on?
2. Reproducibility. If we ask about a figure from March, can you produce it and its inputs?
3. Consistency. Does the number in the monthly letter match the number in the system?

These questions examine the records and controls behind the risk model. Retained inputs and model settings make a figure easier to review. Each allocator decides how that evidence fits its own due-diligence process.

Where TradePolaris fits

Aegis provides VaR and CVaR, forecast volatility, factor attribution, concentration, correlation, drawdown, liquidity and stress scenarios, with an optimiser under name, sector and volatility constraints. Supported option legs and futures positions enter the risk calculations through documented instrument mappings and contract multipliers. Review the coverage and assumptions beside the result.

The risk-limit arithmetic is public, and portfolio calculations retain their method and coverage. Missing history affects checks differently. Decision Review withholds checks that require complete modeled coverage. A standing monitor can evaluate its modeled subset and states what it measured. Inspect that distinction before comparing a proposed decision with a monitoring result.

The exports are a structured CSV meant to be read and a vector PDF with methodology, confidence, concentration, factor, correlation, drawdown, liquidity, options and caveat sections. A reader can see how each figure was produced without asking.