SphinxRisk
Portfolio & risk analytics

Your portfolio, measured properly.

You record what you buy and sell. The tool rebuilds your portfolio day by day and computes risk, return, concentration and valuation — and shows you where every number came from.

Your broker tells you what you have.

Not what you are exposed to.

Same portfolio. Same days. The only difference is how much of it you get to see.

What a broker app shows you

Market value31,722.84
Profit+18,716.49
Return+205.0%

That is the whole picture.
Two of those three numbers only go up.

What this shows, for the same holdings

Market value31,722.84
Profit+18,716.49
Return, time-weighted+205.0%
Worst fall from a peak−43.9%
Annualised volatility32.0%
Return per unit of risk1.51
Worst 5% of days, on average−4.6%
Fat-tail reading8.07
Effective positions, of 32.44
Correlation, calm → stressed0.27 → 0.51

Six of those ten can only be computed by rebuilding the portfolio day by day — which is the work, and the reason a broker app does not show them.

Measured on AAPL, MSFT and NVDA, 2015–2024, by the engine behind this page. Neither column is a prediction and neither is a promise of returns: they describe one specific past. The point is not that the right-hand column is better news — a maximum drawdown of 43.9% is worse news — it is that you were exposed to it either way, and only one of the two columns told you.

No number here is written by an artificial intelligence.

The engine calculates, the model explains. Every figure comes out of deterministic functions covered by 315 automated tests. The copilot can only call them and put them into words, and under each answer you see exactly which ones it called.

Not opinions. Sixty years of published method.

Mean-variance optimisation and the efficient frontier (Markowitz, 1952). The reward-to-variability ratio (Sharpe, 1966). Concentration measured by the inverse Herfindahl index. Historical and conditional value at risk. Purged walk-forward validation, so a model is never scored on data it could have seen. These are the standard tools of institutional risk desks, implemented in full and named where they are used, so you can go and read the paper.

Metrics computed
30
risk, return, valuation, regime
Automated tests
315
every figure covered
Years of fundamentals
10
read from SEC filings
Cost to start
0
no card, no broker link

Why you can trust it

Every figure is traceable. You see which tool computed it and with what arguments. There is an automated test that fails any answer containing a number with no origin — not a promise, a check that runs.

Every result carries its caveat attached. Which window it was computed over, how far the price sits from the accounting period it is being compared against, what it assumes and what it does not. A warning that does not travel with the number is a warning nobody reads.

Fundamentals come from the regulator. Not from a reseller: we read the filings companies submit to the SEC themselves. Ten years of history, from the original source rather than a third party's interpretation.

What you will see

Your portfolio, day by day

Rebuilt from your own trades. Deposit €5,000 and your portfolio rises by €5,000 — which is not a gain: returns are time-weighted. The naive calculation is not approximate, it is wrong.

Allocation drift

Each position's weight over time. A winner takes up more and more room without you buying anything: a portfolio concentrates itself, in the direction of whatever already went up.

Simulate before you buy

What a trade would do to your risk. The first thing you see is its correlation with what you already hold, because that is what decides whether you are diversifying or just doubling a bet you already placed.

Fundamental analysis that means it

Margins, cash cycle, ROIC, leverage and multiples. With red flags that cross the income statement against the balance sheet and against cash, and give both readings: the benign one and the worrying one.

A copilot that does not calculate

Ask in plain language. It queries the same engine as everything else, and shows you what it queried.

Three real findings. Two are uncomfortable.

Measured over ten years of AAPL, MSFT and NVDA with this engine. Not examples invented for this page.

Correlation between holdings
0.27 → 0.51
in calm, and under stress
Sharpe of the optimal portfolio
3.40 → 0.77
once regime risk is counted
Drawdown model hit rate
1 in 5
correct warnings. We say so.

Diversification weakens exactly when you need it. A single covariance matrix averages that effect away until it disappears.

The risk is not where Markowitz looks. Not knowing which regime you are in is itself risk, and a classical model cannot represent it.

The prediction model does not work. It ships with that number beside it instead of being tuned until it looked good. Absence of signal is a finding too.

What it does not do

It is not a financial adviser and it will not tell you what to buy. It places no orders and connects to no broker to trade. It promises no returns: everything it computes describes one specific past, and nothing guarantees the future will resemble it.

It can analyse what a decision would imply. It cannot make the decision for you — and any tool that claims otherwise is selling you a certainty that does not exist. The full limits, here.

Upload your broker's CSV and have the analysis in a minute.

English and Spanish headers are both understood. Nothing is saved until you confirm it.