The question behind the question
Firms evaluating attribution ask which model the system runs. Brinson‑Fachler or Brinson‑Hood‑Beebower. Which fixed income decomposition. How currency is handled. These are fair questions and we are happy to answer them, but they are rarely the question that decides whether the system changes anything.
The question that decides it is simpler: who sees the output, and when? A faultless decomposition that reaches the portfolio manager on the twelfth working day is a post‑mortem. An honest one on their screen before the market opens is a portfolio construction aid. The maths is identical. The value is not.
We built this system after years on the other side of that divide — sitting with portfolio managers, watching attribution arrive too late and too narrowly distributed to change a single position. Everything below follows from trying to fix that.
Accuracy is the entry ticket, not the argument
Every vendor claims accuracy, so the claim carries no information. What carries information is where the numbers come from.
A system that takes security‑level returns from a third‑party engine and decomposes them inherits whatever that engine produced, including its errors. Top‑line performance can look entirely correct while individual security returns are wrong — the errors net out at the fund level and reappear, loudly, the moment you try to attribute them. Attribution built on that foundation spends its first week of every month being reconciled rather than read.
We recalculate performance from your transactions, aligned to your own ABOR or IBOR. Not because recalculation is clever, but because it is the only way to find the error before it reaches the attribution, and because it changes the tone of the conversation. Portfolio managers stop asking whether the numbers are right and start asking what they mean.
There is no residual in our output, and that is not a tolerance setting. Every effect is derived from the same transaction‑level return calculation that produced the return itself, so the effects sum to the active return by construction. There is no smoothing term because there is nothing left over to smooth.
The models have to fit the portfolio, not the other way round
Portfolios are not what they were when the standard frameworks were written. Overlays, hedges, pooled vehicles, illiquids and multi‑asset structures all sit alongside the long equity and bond positions the classical models assume. A single rigid model applied to all of it produces output that is arithmetically correct and practically useless.
So the Brinson family extends into multi‑level, multi‑benchmark form, with allocation and selection attributed separately at strategic and tactical layers rather than collapsed into one number that nobody owns. Fixed income splits into carry and roll‑down, curve shift, curve twist, spread and sector, issue selection and currency — the effects a rates desk already talks in. Currency is modelled through exposure assets and synthetic hedging instruments rather than security by security, which keeps the portfolio and the benchmark on the same hedging basis and makes it obvious whether an outcome came from the local asset, the currency move or the way the hedge was implemented.
Your classifications, your analytics, your benchmarks
The system runs inside your own Azure tenancy, on your prices, your sector and rating classifications, your durations and your analytics, under the agreements you already hold. Nothing is redistributed and nothing leaves your ownership, so there is no new licence to negotiate and nothing new to pay for.
The licensing is the dull half of the reason. The interesting half is that when a portfolio manager opens the exposure column, the duration they see is the duration they have been quoting all morning. Attribution built on a vendor's own analytics is correct against a world the investment team does not live in, and every disagreement with it has to be argued rather than read.
Data quality is a by‑product, not a project
The usual objection to using a client's own data is that it will be worse than a vendor's. In our experience the opposite holds, and the reason is structural rather than flattering: internal data is already feeding risk systems, dealing systems, client reporting and the front office itself. Dozens of people have already looked at it with a reason to care. The errors that survive that are rare, and the ones that do not have already been fixed by someone else, for their own reasons, before attribution ever sees them.
Clients tell us this is what they experience. A separate attribution data set, maintained by one team for one purpose, has no such immune system.
What “at the desktop” actually requires
All of the above is necessary and none of it is sufficient, because a system can be accurate, well modelled and correctly sourced and still be seen by four people. Three things decide whether it reaches the desk.
- It has to be daily. Monthly attribution explains; daily attribution informs. Performance teams can find a discrepancy overnight instead of at month end, and a portfolio manager can see what a position did while there is still something to be done about it.
- It cannot be priced per seat. Per‑seat licensing makes every additional reader a cost to be justified, so firms ration access to the people who have to have it. We charge per portfolio, at $100 per portfolio per month, with unlimited users — deliberately, because the more people looking at the results, the more the results are worth.
- It has to be self‑service. Mapping, re‑runs and new portfolios belong to the team that needs them. The moment a change requires a vendor ticket, the system stops being a tool and goes back to being a report.
The same three decide how long implementation takes, which is why ours is measured in weeks rather than the months or years the industry is used to — four weeks from signature to first signed‑off month, and that month is free.
The test
Whatever you are running now, four questions will tell you where it stands:
- Does it reconcile to your official performance without manual help?
- Does it reflect how the portfolio is actually constructed, or how the system needs it to be?
- Does it arrive early enough to change a decision rather than explain one?
- Can everyone who would benefit from it actually open it?
Most firms we talk to can answer the first two. The third is harder. The fourth is almost always a pricing answer dressed as a technical one.