Learn / Advanced ALM

Economic Scenario Generators

A deterministic projection is one weather forecast. An ESG writes many. One hundred annual Euler paths of Vasicek, CIR, or Hull–White discount the same classroom annuity. A correlated equity sleeve rides beside the short rate. This is not a calibrated vendor ESG and not a prescribed C3 scenario set.

01

Deterministic versus stochastic

A single discount rate answers “what if this path.” Monte Carlo answers “what is the distribution.” Cheap-rate paths make the annuity expensive; dear-rate paths make it cheap. High present value is the writer's loss, so the 95th percentile of liability PV is the expensive tail.

Analogy: one rainfall forecast versus a hundred. The mean tells you how much grain to plant. The expensive (high-PV) tail tells you whether the writer can fund the guarantee when rates stay cheap.

Vasicek and Hull–White let the short rate wander and pull it back toward a mean. CIR does the same but will not go below zero. Equity is a separate sleeve that can fall when rates rise if you set a negative correlation.

02

One hundred short-rate paths

Teaching lab

One hundred short-rate paths

Annual Euler steps for 20 years. Each path discounts the same $8 million classroom annuity. The seed is fixed, so a refresh does not invent a new fan. Not a calibrated ESG.

Mean liability PV
$110,107,225
5th percentile (cheap tail)
$97,181,161
95th percentile (expensive tail)
$124,990,722
Mean equity sleeve
$61,696,581
Equity 5th / 95th
$10,123,497 / $167,907,797
Sample short-rate pathsTwenty of the simulated short-rate paths.Sample pathsTeaching sketch
Gold is the first path. The others show the fan. CIR is floored at zero. Hull–White here uses a constant reversion level — a classroom stand-in for a fitted θ_t.
Liability PV histogramDistribution of the classroom annuity present value across paths.Liability PV histogram
Wider σ or a higher path count changes the histogram. Mean and tails are the validation numbers a desk would compare to a target. Synthetic.

A deterministic projection is one weather forecast. An ESG writes many. Cheap-rate paths make the annuity expensive; dear-rate paths make it cheap. High present value is the writer's loss, so the 95th percentile of liability PV is the expensive tail.

03

Behind the scenes

  1. Draw a unit normal with a fixed seed so a refresh does not invent a new fan.
  2. Step the short rate once a year for 20 years under the chosen model.
  3. Pair an equity shock at correlation ρ and grow a $20 million sleeve as geometric Brownian motion.
  4. Discount the $8 million classroom annuity on each rate path.
  5. Rank the present values. Report mean, 5th, and 95th percentiles, and the equity fan.

04

Curves, calibration, and real-world versus risk-neutral

A production ESG starts from a yield curve: spot rates, forwards, and a discount curve. Calibration fits mean reversion and volatility so that the generator prices (or at least recovers) today’s curve and a volatility target. This lab skips that solve. Spot, forward, and discount construction stay on the yield curves lesson. Hull–White as a fitted short-rate model stays on Hull–White.

Economic / real-world scenarios are for planning, capital, and cash-flow testing: they can carry a risk premium. Risk-neutral scenarios are for pricing traded or embedded options: the drift is whatever keeps traded assets fair. Mixing the two is a common model-validation finding. Tail scenarios are the outer percentiles of the histogram. Adding paths is a crude variance reduction; a production shop would also use antithetics, control variates, or stratified sampling.

Scenario correlation here is one number ρ between the rate shock and the equity shock. A multi-factor ESG would add credit, inflation, and a full correlation matrix. Those factors are named, not simulated.

05

A real-world ALM use

A stochastic ALM run asks how wide the liability PV fan is at 100 versus 200 paths, and whether a higher rate volatility or a more negative equity correlation fattens the surplus tail. Raise σ, then raise path count, and watch the histogram and the standard teaching percentiles. Related typed-versus-drawn paths live on scenario generation.

06

Limitations

Annual Euler steps, one factor, a constant Hull–White θ, no credit, no inflation, no prescribed statutory scenario set, and no curve calibration. The equity sleeve is GBM with a constant ρ. Percentiles from 100 paths are noisy. Do not treat the fan as a C3 or ORSA result.

Related pages: Advanced ALM hub · Advanced quantitative finance · Sensitivity & stress testing.

Trades | Learn. Practice. Analyze. Improve.