Optimisation in a spreadsheet
A spreadsheet solver rarely states which weight bounds made the problem infeasible, or which covariance estimator it assumed — an analyst is left explaining a number they cannot fully trace.
For portfolio and risk analysts
From a spreadsheet allocation to one you can defend.
An optimiser that prints what it assumed, a stress test replayed on the dates it actually happened, and a report a committee can read without a follow-up call to ask what the numbers mean.
01A week with Quant Compose
From a universe and its risk diagnostics to a report a committee can act on, across one working week.
Day 1
Universe and risk diagnostics
Apply a universe and compute the full risk and return set — annualised return, volatility, Sharpe, Sortino, Calmar, drawdown, beta and correlation — for every security before any allocation decision is made.
Market Explorer Growth of one unit for every security in the universe, rebased to the first shared date. Day 2
Choose a covariance estimator
Sample covariance, Ledoit-Wolf shrinkage, a denoised random-matrix-theory estimator, or EWMA at a 0.94 decay. The estimator is a choice the analyst makes explicitly, not a default inherited silently.
Risk & Return Correlation matrix of daily returns across the universe, the input every covariance estimator starts from. Day 3
Optimise, with refusals recorded
Mean-variance, minimum variance, maximum Sharpe, risk parity, hierarchical risk parity, Kelly or CVaR optimisation. An infeasible weight-bound combination is refused outright, and frontier points the solver could not reach are listed with the specific reason rather than dropped.
Portfolio Optimisation Max-Sharpe, minimum-variance and equal-weight allocations side by side. Day 4
Stress the result
Replay one of six real historical episodes — the 2008 crisis, the 2010 flash crash, the 2011 European sovereign selloff, the 2015 China devaluation, February 2018, or the 2020 Covid crash — on the portfolio's own loaded price history, or propagate a hypothetical shock through each asset's beta.
Scenario & Stress Testing Loss contribution by asset under the shock, beside base-versus-stress risk comparison. Day 5
Report to the committee
Export the fixed evidence pack in PDF, DOCX, HTML or JSON, with portfolio construction and benchmark-relative risk as their own sections and the no-live-order-routing statement printed in the footer of every page.
Evidence pack (PDF) Page 1 of the exported evidence pack: universe and data, absolute risk and return, benchmark-relative risk, portfolio construction.
02Three problems it removes
Named specifically, not as a category.
Ad hoc stress tests
A stress scenario built from scratch each time is a new set of assumptions each time, and a shock magnitude picked without a stated method is hard to defend when questioned.
Manual reporting
Assembling charts and tables into a committee memo by hand costs a day that could go to the analysis itself, and invites a transcription error the source data never had.
03Assumptions, printed
Method that holds up when someone checks.
A committee does not need to trust the analyst's memory of what was assumed — the assumptions are in the output.
- Equal weight as the standing baseline
- Equal weight is always available as the baseline optimisers frequently fail to beat, so a fancier allocation has to earn its place against it.
- Refused frontier points, named
- A point the solver could not reach under the stated constraints is recorded with the reason, never silently dropped from the chart.
- The covariance estimator is a stated choice
- Sample, Ledoit-Wolf, denoised or EWMA — whichever is used is named in the output, not buried in a default nobody set deliberately.
- Losses reported as positive numbers
- Value at Risk and Conditional VaR are reported as positive loss figures across all four VaR methods, avoiding the ambiguity of a signed convention that reverses between reports.
- VaR methods named
- Historical, parametric, Cornish-Fisher and Monte Carlo are each labelled explicitly rather than presented as one undifferentiated "VaR" figure.
04Objections, taken seriously
The questions this audience actually asks.
Can I trust the optimiser's assumptions?
They are printed. Frontier points the solver could not reach are listed with the reason, equal weight is always the baseline, and the covariance estimator is a choice you make, not a default you inherit.
Which VaR?
You choose from four named methods — historical, parametric, Cornish-Fisher and Monte Carlo — and the choice appears in the output rather than being averaged away.
Is it execution?
No. The workspace is a research and analytics tool; it does not place or route an order, run against a broker, or manage live positions.
Can I use my holdings?
Upload your own OHLCV price history as a CSV — the accepted format is date, ticker, open, high, low, close and volume — and build the allocation from it using any of the seven portfolio construction methods. There is no separate holdings-file import; weights come from a construction method run on your data.
05In beta, for you
What is available to portfolio and risk teams today.
- Full access to the seven construction methods, four covariance estimators and four VaR methods
- All six historical stress episodes plus hypothetical shock testing, at no cost during the beta
- PDF, DOCX, HTML and JSON export of the same evidence model, so a committee memo and a downstream system read the same numbers
- Direct feedback into which additional stress episodes and estimators get added next
Other audiences, and further reading
Related capabilities
Build the allocation you can defend in the room.
Beta access is open to individual analysts today; teams standardising across a desk should get in touch directly.