Advanced Finance

Stochastic Cash-flow Modelling with Correlated Drivers

Replace a base case and three sensitivities with a distribution built from drivers that move together, and get the one number a deterministic model can never produce: the probability of breaching the covenant.

  • Expert
  • 15 min total
  • 14 chapters

What decision this helps you make: Whether the base case in front of you describes a likely outcome or an arithmetic artifact — and how much of the risk in a business comes from its drivers moving together rather than from any one of them being wrong.

What this topic is

Stochastic cash flow modelling replaces point estimates of the drivers of a business — price, volume, cost, availability, exchange rate — with probability distributions and stochastic processes, imposes a dependence structure between them, and then simulates the model many thousands of times to produce a distribution of outcomes rather than a single number. The correlation part is what distinguishes it from naive simulation: drivers rarely move independently, and the joint behaviour of price and volume, or of input cost and output price, usually matters more to the tail of the distribution than the volatility of either driver on its own.

Why it matters

A deterministic model with a base case and an upside and a downside cannot answer the questions that actually determine whether a financing works: what is the probability of breaching a covenant, what is the tenth percentile of cash flow, how much of the downside comes from two things going wrong together. Worse, a model run at mean inputs does not generally produce the mean output, and where any part of the structure is non-linear — a cash sweep, a coverage test, a tax loss carried forward, an option to abandon — the base case can sit a long way from the middle of the distribution it is meant to represent. In the worked example in this lesson, the base case gives an EBITDA of $18.0M and the median outcome is $16.4M, with a 29% chance of falling below the coverage covenant in any given year.

Who should learn it

Project and infrastructure financiers, corporate development and planning teams, risk managers who own the downside case, and anyone who has ever been asked how confident they are in a forecast and has had to answer with an adjective.

What you will understand

  • How to choose a process that matches each driver's economics rather than defaulting to a normal distribution
  • How to impose correlation with a Cholesky factor, and what to do when the correlation matrix is not valid
  • Why the deterministic base case is neither the mean nor the median of the distribution it summarises
  • How many paths you actually need, and why a tail statistic needs far more than a mean does

Prerequisites

Common misconception

"Running a Monte Carlo makes the forecast more accurate." It does not make the forecast more accurate. It makes the uncertainty in the forecast explicit and forces you to state it in a form somebody can argue with, which is genuinely valuable and completely different. A simulation is exactly as good as its input distributions and its dependence structure, and the dependence structure is the input with the weakest empirical foundation and the largest influence on the tail. A fan chart produced from a correlation matrix estimated over a calm decade is a precise statement about a world that did not include the event you are worried about.