## Don't shoot the messenger

Stochastic projection models have many advantages — they not only give best-estimate projections, but also confidence intervals around those projections. Mortality improvements have been accelerating for many years, so extrapolative models naturally project continued improvements, or even continued acceleration in improvements. This is in contrast to targeting models, which at the time of writing "predict" a deceleration in the rate of improvement.

But what should managers do when their actuaries produce projections which result in stronger bases than others?  Stronger bases mean higher pension reserves, or more expensive annuities.  It is understandable that a pensions actuary will hesitate before recommending something which will increase reserves and lead to an awkward conversation with a client.  Bearers of inconvenient predictions can have a difficult time of it, although few are quite as unlucky as Nikolai Kondratieff:

"Nikolai Kondratieff was executed on Stalin's orders in the mid-1930s because his econometric model predicted, accurately as it turned out, that collectivization of Russian agriculture would lead to sharp decline in farm production."

Peter F. Drucker, Innovation and Entrepreneurship, p4.

According to Drucker, Kondratieff's model made an accurate prediction, but one for which he was not thanked.  In contrast, Trofim Lysenko's pseudo-science about the inheritance of acquired characteristics and agriculture were more like what Stalin wanted to hear, and Lysenko thrived as a result.  However, the effects of ignoring inconvenient predictions, and preferring answers which fitted beliefs, not facts, led to catastrophe for Soviet agriculture and even famine.

What lessons does this all have for mortality projections?  Firstly, the Kondratiev point — managers and clients should not dismiss inconvenient projections, but should think very carefully about why extrapolative models, including many stochastic models, project strong rates of mortality improvement.  These models are simply telling you that the near future is going to be a continuation of the recent past.  This may prove to be inconvenient in the short term, but as the Soviets found out, the results of only choosing predictions which fit your beliefs can be disastrous.

Secondly, the Lysenko point — we must beware of listening only to what we want to hear.  If a projection conveniently allows for lower reserves or prices, it should be benchmarked against a peer-reviewed stochastic model to make sure it isn't too optimistic.

Assume we have a random variable, $$X$$, with expected value ... Read more