The CIP review that finds nothing: what to test beyond fund level tolerances

Most annual reviews of a centralised investment proposition conclude that the proposition is working. I would treat that as mildly uncomfortable rather than reassuring, because in my experience it is usually a function of what was tested.

A typical review asks three things. Is each holding still within its tolerance. Are the managers still in place and the mandates unchanged. Is performance broadly in line with the comparator. Those are sensible questions and the answer is nearly always yes. What they establish is that nothing has broken.

Nothing having broken is genuinely worth knowing. It is not the same as the proposition still fitting the clients who are in it.

The tests that move a decision

The questions that change what a firm does next are about dispersion rather than tolerance.

One: how far apart are clients in the same model. Two clients placed in the same balanced model eighteen months apart do not hold the same thing, and after a period of drift and different contribution patterns they may not be close. Measure the spread of current asset allocation for everybody in one model. A wide spread tells you the rebalancing discipline is not doing what the documentation says it does.

Two: how many clients sit outside the profile on their own file. Not outside the model's mandate, outside the risk profile recorded at their last fact find. This is the number that a file review would find, and it is far more useful than a fund level screen because it is client level and it is the thing a client would recognise as wrong.

Three: how much of the range is actually used. A nine model range where three models hold most of the assets is a range with nine models of governance and three in service. That is a cost, a documentation burden, and a set of small portfolios where trading is inefficient.

Four: where the outcomes sit relative to what the client was told to expect. Not relative to a benchmark, relative to the planning assumption the advice was built on. A model can be comfortably in line with its comparator and still be some way from the return the cashflow plan assumed.

A worked example

Take a firm with 212 clients across a five model range.

None of those four findings would have come out of a fund level screen, and each one produced an action. Compare that with a review that confirms all twenty three underlying funds remain within tolerance, which produces a document.

What this is not

It is not an argument for changing the proposition. Three of the four findings above led to a change in process rather than in holdings, and that is the usual pattern. A review that keeps concluding "replace the fund" is probably chasing performance.

It is also not an argument against outsourcing. A firm using a discretionary manager has the same four questions, it just asks some of them of the manager and needs the data to do so. The client level tests, profile drift and range usage, remain the firm's own responsibility whoever runs the models.

The part that is hard

All four tests need client level data, holdings joined to profiles joined to models, and that is precisely the join most firms cannot do quickly. Which is worth saying plainly: the reason these reviews stay at fund level is often not that nobody thought of the better questions. It is that the better questions take a week to answer and the fund screen takes an afternoon.

That is a data problem rather than an investment one, and it is worth solving, because a review that cannot find anything will keep on not finding anything.