Skip to content

Correlations ​

By default each assumption is drawn independently of the others. In reality many variables move together:

  • If the price rises, the quantity demanded tends to fall (negative correlation).
  • If inflation rises, costs and prices rise together (positive correlation).
  • Local and international equities tend to move together.

Ignoring these relationships can under- or overstate risk. For instance, if price and quantity are negatively correlated, “high price and high quantity” scenarios are rarer than independence would imply, and revenue is less dispersed.

Defining correlations ​

  1. Define at least two enabled assumptions.
  2. Click Simulation → Correlations (Alt+C).
  3. A matrix with every assumption appears. Type coefficients in the upper triangle; the matrix is symmetric, so the lower triangle fills itself. The diagonal is always 1.
  4. Click OK.

Dialog buttons:

  • Set all to 0: removes every correlation.
  • Adjust to nearest: appears when the matrix is invalid (see below).

The Model explorer footer shows how many correlations are defined.

What the coefficient means ​

Coefficients are rank (Spearman) correlations between −1 and 1:

ValueInterpretation
+1Perfect increasing relationship: when one is at its highest, so is the other.
+0.5They tend to move in the same direction (moderate).
0No relationship (independent).
−0.5They tend to move in opposite directions.
−1Perfect decreasing relationship.

Rank correlation (rather than Pearson) is used because it works with any pair of distributions — a Normal with a Triangular, a Poisson with a Lognormal — without changing the shape of either.

Where do I get the coefficient?

  • From historical data: compute the correlation between the two series (e.g. with =CORREL(…) or the regression tool).
  • From the literature or market studies (price elasticity of demand).
  • From expert judgement: −0.3 weak, −0.5 moderate, −0.8 strong. Justify it in your report and test how sensitive the result is to it.

How they are applied: the Iman–Conover method ​

OpenRiskSim induces correlations with the Iman and Conover (1982) method, the same approach used by Risk Simulator, @RISK and Crystal Ball:

  1. Each assumption's samples are generated independently (exactly following its distribution).
  2. The samples are reordered so that their ranks have the requested correlation.

Because they are only reordered, each assumption keeps exactly its distribution (same mean, same percentiles). The achieved correlation is very close to the requested one but not identical: with 5,000 trials the difference is usually in the second decimal.

Non-positive-definite matrix ​

Not every combination of coefficients is possible. For example, if A and B have +0.9 correlation, B and C +0.9, but A and C −0.9, the matrix is inconsistent (if A and C both move with B, they can't move against each other). Mathematically, the matrix must be positive definite.

The dialog tells you:

  • ✅ “Valid matrix (positive definite).”
  • ⚠️ “The matrix is not positive definite: the engine will use the nearest valid matrix.”

In the second case you can click Adjust to nearest to see and keep the corrected matrix (computed with Higham's nearest-correlation-matrix algorithm). If you don't, the engine corrects it anyway when simulating.

Turning them on or off ​

In Simulation → Settings, the Apply correlations checkbox lets you run with or without them without deleting the matrix. Useful to show in your report how the risk changes when correlation is considered.

Example: price and demand ​

The “Project evaluation” example defines a −0.5 correlation between Unit price and Units sold year 1. To see its effect:

  1. Run with the original settings and note the NPV's standard deviation and the probability of NPV ≥ 0.
  2. In Settings, untick Apply correlations and run again.
  3. Compare: without the negative correlation, extreme scenarios (high price and many units, or low price and few units) are more frequent, so the NPV's dispersion increases (with 10,000 trials the standard deviation goes from about $70,400 to $97,200).

WARNING

Correlations only apply between enabled assumptions. If you disable an assumption, its correlations are ignored.

Released under the GPL-3.0 license · © 2026 Fernando Daza and contributors