Why lognormal?
If continuously compounded returns are normally distributed, then the price, which is the starting price times e raised to the return, follows a lognormal distribution. Two features make it a good fit for prices:
- It is bounded below by zero. A stock can lose everything but not go below zero.
- It is skewed to the right. There is no upper limit, so the distribution has a long right tail.
The normal distribution, by contrast, runs from minus to plus infinity, which suits returns but not prices.
Continuously compounded returns
The continuously compounded return between two prices is:
r = ln(P1 / P0) = ln(1 + HPR)
Its useful property is that it is additive over time: the return over several periods is the sum of the period returns. Monthly returns of 2%, −1% and 3% (continuously compounded) give a quarterly return of 4%. If monthly returns are independent and identically distributed, the mean scales with time and the variance scales with time, so volatility scales with the square root of time:
σ(annual) = σ(monthly) × √12
Monte Carlo simulation
Monte Carlo simulation estimates the range of possible outcomes when the answer is too complex to work out with a formula. The steps:
- Choose the variables that drive the result (for example returns, inflation, withdrawals).
- Choose a probability distribution for each, with parameters such as mean and standard deviation, and any correlations between them.
- Use a computer to draw random values for each variable and compute the outcome.
- Repeat thousands of times.
- Look at the distribution of outcomes: average, percentiles, probability of failure.
Uses include valuing complex securities, testing retirement plans and estimating risk measures. Strengths: it handles complexity and gives a full distribution, not a single number. Weaknesses: it is only as good as the assumed distributions, and it is a statistical estimate rather than an analytic answer, so it does not explain cause and effect.
Try our own Monte Carlo simulator to see the method in action.
Historical simulation
Historical simulation draws scenarios from actual past returns instead of an assumed distribution. It captures real-world features like fat tails without having to model them, but it assumes the future will resemble the sample period, and it cannot produce events that did not happen in that period.
Bootstrapping
Bootstrapping treats an observed sample as if it were the whole population. It draws many new samples with replacement from the original sample and computes the statistic of interest (such as the mean) for each. The spread of those results estimates how uncertain the statistic is, without assuming a theoretical distribution.
| Method | Draws from | Main strength | Main weakness |
|---|---|---|---|
| Monte Carlo | Assumed distributions | Flexible, can model any scenario | Depends on assumptions |
| Historical simulation | Actual past data | No distribution assumption | Limited to past events |
| Bootstrapping | The sample itself, with replacement | Estimates uncertainty of a statistic | Sample must represent the population |
Exam focus
Know why prices are lognormal and returns normal, that continuously compounded returns add up, how to scale volatility by the square root of time, and the difference between Monte Carlo, historical simulation and bootstrapping.