Volatility v. Trend Risk

The year 1992 was important in the development of forecasting methods: Ronald Lee and Lawrence Carter published their highly influential paper on forecasting US mortality. The problem is difficult: given matrices of deaths and exposures (rows indexed by age and columns by year) can we forecast future death rates?  Lee and Carter designed a model specifically to solve this problem:

log μx,y = αx + βxκy        (1)

where αx measures the average mortality at age xκy measures the effect of year y; this year effect is modulated by an age dependent coefficient, βx.  Lee and Carter used US data up to 1989 and here I’ve followed them by using data on US males aged 60–90 between 1933–1989, downloaded from the Human Mortality Database.

null
Figure 1. Fitted κ for male US data, years 1933–89.

Now we see what they were up to.  If we assume that the αx and βx are stable in time then all we need to do is forecast the κy in Figure 1. The model has reduced forecasting a table to forecasting a one-dimensional time series.  Clever!

Now comes the tricky bit.  What time series do we choose?  Lee and Carter argued strongly from a plot similar to Figure 1 that the forecast should be based on a random walk with drift.  Thus:

κy = κy-1 + a + εy,   εy ˜ N(0, σ2),   y = 1934, ... , 1989          (2)

where the drift and volatility parameters a and σ are to be estimated from the values of κy in Figure 1.  We write this expression in terms of dy = κy - κy-1:

dy = a + εy,   εy ˜ N(0, σ2),   y = 1934, ... , 1989.

Our estimate of the drift parameter a is thus the mean of dy, about −0.008 for these data; the estimate of σ2 is the sample variance of dy, which gives a standard error of about 0.003 for our drift parameter.  A very rough 95% confidence interval for a is (−0.014, −0.002) so a is not precisely determined.  This has alarming consequences for the accuracy of our forecast.  Figure 2 shows the best-estimate forecast for κ together with a 95% confidence interval:

null
Figure 2. Fitted and forecast κ, data 1933–89, forecast to 2048.

Future sample paths generated by model (2) depend on both parameter risk and stochastic risk.  We can isolate stochastic risk and Figure 3 shows one hundred sample paths simulated under the assumption that the long term trend is fixed at the central forecast.  The dashed green line is the 95% envelope, i.e. the probability that the envelope contains a sample path is approximately 0.95.

null
Figure 3. Sample paths with stochastic error only.

Figure 4 shows one hundred sample paths when only parameter uncertainty in the drift parameter is taken into account:

null
Figure 4. Sample paths with parameter error only.

while Figure 5 shows one hundred sample paths when we allow for both  parameter and stochastic risk:

null
Figure 5. Sample paths with stochastic and parameter error.

Figure 6 summarizes the relative contributions of parameter and stochastic risk:

null
Figure 6. Central forecast with various sample path 95% envelopes.

Here are a few of the many comments that can be made on this exercise:

  • Volatility is the dominant source of error in the short term.
  • The importance of long term trend gradually increases over time until it eventually overtakes volatility as the “senior partner”.
  • The mean-reversion property of the random walk with drift is obvious in Figure 3.  Are you comfortable that the course of future mortality behaves in this way?  I merely ask the question.

And a final question: is there an elephant in the room?  The above discussion assumes that the model is correct; there is no way we can be sure of this.  A cautious actuary will consider a range of models.  Only in this way can he or she come to an appreciation and understanding of longevity risk.

You can find some additional material in support of this blog in the technical note referenced below.

References:

Lee, R. D. & Carter, L. R. (1992) Modeling and forecasting US mortality. Journal of the American Statistical Association, vol 87, 659–675.

Currie, I. D. A technical note on estimating and forecasting with the random walk with drift.

Parameter uncertainty in the Projections Toolkit

All standard-error sheets produced by the Projections Toolkit automatically include parameter uncertainty.  The standardised spreadsheet layout of the output enables this uncertainty to be automatically built in to your calculations, regardless of the underlying structure of the model chosen.

Most models also support the generation of sample paths.  The sources of uncertainty are independently switchable: you can have parameter uncertainty only, volatility only, or both risks combined in your sample paths.

Previous posts

Between a rock and a hard place

The Advocate General of the European Court has recently opined that "the use of actuarial factors based on sex is incompatible with the principle of equal treatment for men and women".
Tags: Filter information matrix by tag: gender, Filter information matrix by tag: Solvency II

Putting the TAS into Longevitas

In the UK the Board for Actuarial Standards (BAS) has published a series of Technical Actuarial Standards (TAS).  At the time of writing there are three standards which are either in force or shortly will be: one for data (TAS-D), one for reports (TAS-R) and one for actuarial models (TAS-M).
Tags: Filter information matrix by tag: BAS, Filter information matrix by tag: TAS

Add new comment

Restricted HTML

  • Allowed HTML tags: <a href hreflang> <em> <strong> <cite> <blockquote cite> <code> <ul type> <ol start type> <li> <dl> <dt> <dd> <h2 id> <h3 id> <h4 id> <h5 id> <h6 id>
  • Lines and paragraphs break automatically.
  • Web page addresses and email addresses turn into links automatically.