What is “Portfolio Management”?

Portfolio Management (PM) is extracting greater value from an existing portfolio of insurance policies by identifying opportunities for growth or margin improvement. 

The need for PM, in brief, is driven by the twin observations (a) that every portfolio has untapped potential if it can only be found; and (b) that alternative means to achieve profit or growth involve risky and expensive activities like acquisitions.

More specifically, we define PM as follows:

Portfolio Management: PM is a process where human decision-making is informed by analytical insight to derive greater value from an existing portfolio.

The process is a control cycle: it repeats regularly with reassessments and refinements.  Emerging data is ingested each iteration, and previous assessments and actions are tested and updated:

Let’s parse the definition, in reverse order:

A.    Analytical insight

There is a lot of analytical heavy-lifting for the richness of insight to be able to identify and justify profit-enhancing portfolio decisions.

Backwards

We need credible historical results – with justified assessment of IBNR – for a view of the future is always informed by the past.

Forwards

But that view of the future is the most important aspect of the analysis.  This is portfolio gold dust, and is summarised in what I call Emerging Profitability.  This is the expected profit margin in the business being written now (and in the near future).  It’s informed by the past but also takes account of portfolio-specific changes and market context.  This might be rate actions we’ve taken, past re-shaping exercises like partial exits, or wider issues like inflation, intensity of competition and the like.

No silos

In traditional actuarial terms, what PM needs is neither reserving alone nor pricing alone.  It needs the best of both.  The typical reserving process looks backwards and does not take full account of current underwriting conditions (and, besides, is usually geared around accounting and solvency principles of prudence).  The pricing assessment tends to magnify any biases of model builders (or model users, or their data), and can lose sight of the portfolio wood in seeking the perfect policy-price trees. 

The two combined can hold each other to account and deliver credible Emerging Profitability assessments where either in a silo would fail.

Allocation, allocation, allocation

We’re assessing IBNR and Emerging Profitability, the former usually by accident year (AY) and the latter by underwriting year (UY).  But the two inter-depend, so each must be allocated to the other’s cohort.  We must be able to glide seamlessly between AY and UY, a fiddly exercise if performed from scratch each time so best embedded in our process without the need for revisiting.

Similarly, all results must be broken down into meaningful chunks.  “Improve portfolio loss ratio by 1%”, while laudable as an overall objective, isn’t really an action of any use.  But “20% increase in price for properties with high flood risk scores” or “reduce the 27% commission for broker X to 23%” are specific and targeted, and make sense as actions.  This requires another type of allocation, from the aggregated level at which actuarial assessments are made to more detailed, granular levels where decisions will be made.  As with AY / UY, this allocation exercise must be integrally embedded in the PM process.

B.     Human decision-making

Insurance isn’t easy for many reasons.  Products are complex and manifold; there is a long delay between making decisions and observing their results; the number of distribution and operating models cannot be counted on a millipede’s toes; data is always incomplete for the purposes of identifying “the truth” and often patchy, volatile or wrong.

As a field of finance whose key currency is data, insurance is obviously a numbers game.  But because these numbers are contextually nuanced, they demand significant judgment and experience to interpret.  And human relationships (especially in distribution models) are the glue that holds the industry together.  So insurance is also a people game.

For me, failure to insert a human check into the chain of “data —> analysis —> outcome —> repeat” would be a huge mistake.  There exist areas of limited scope where pure algorithms may operate to optimise results.  Such work considers the effects of individual rating factors to try to get the most value from a single pricing structure.  But this is limited and can’t catch problems with the pricing structure itself, or compare one pricing model to another, or consider the effects of anything other than pricing.

PM is there in part to hold the pricing models to account.  We cannot rely on machines alone to do this.  The heavy-duty analysis in (A) above is important, but it must feed a process governed and managed by people in charge of making decisions.

The decisions will be portfolio actions like identifying areas for growth, shrinkage or exit; pricing changes; product or risk-selection criteria changes; or commission renegotiation or expense management.

These actions will lead to margin improvement.  And they will facilitate growth either directly or by liberating expenses, resources and management time to deploy elsewhere in the business.

C.     Process

Portfolio management is a process, not an event.

A one-off piece of work to assess the various profitability signals within a portfolio has some value but it is limited.  As noted above, the workload is considerable and failing to make it repeatable would be a dereliction of scale-economy duty.  The need to incorporate emerging data, and identify emerging signals and trends, is an effort that pays dividends over time, not at once.  Tracking the effects of actions previously identified requires repeatability and self-monitoring.  The judgments we make in our analysis need to be monitored and compared with experience which materialises subsequently.  And if this effort has value – and I passionately believe it does! – then who wouldn’t want to turn it into a repeatable process?

By building a process for PM we achieve analytical efficiency, the means for complex models to self-correct, and maximal profitable insight.