Reference Class Forecasting

What is Reference Class Forecasting?

Reference Class Forecasting (RCF) is a conceptually simple, yet powerful technique for predicting project durations and costs. It was developed by leading economists including Amos Tversky and Daniel Kahneman, and its development led to a Nobel Prize in Economics.

This technique responds to some simple observations. We tend to predict outcomes based on:

• what we know (which is often limited, despite our experiences), and
• what we hope (thereby exhibiting aspiration or optimism bias).

RCF shares similar aims to other estimating techniques, but it is the focus on ensembles and probabilities that makes it so powerful, and worth having in our estimating and prediction toolbox.

How does it work?

When we wish to estimate a new project, we first work out what type of project it is. For example, is it a construction project, or a software engineering project? If construction, what type of construction - infrastructure, commercial buildings, transport, housing etc.

Then, having determined the class, we seek past examples of completed projects in that class. We record the costs and durations of each, and develop a profile. What is the range of durations and costs? How tightly grouped are they? Are there significant outliers?

We use this profile to estimate our project and, crucially, provide some bounds to the low and high scenarios.

The Challenges with RCF

It’s that simple! Or so we’d like to think. The challenges are many, not least getting access to good quality data, and interpreting that data in a way that is meaningful, statistically valid and that read across to your project.

RCF is not unique in using the techniques of probabilities, or distributions. Parametric estimating can also use probability distributions. Project Science Ltd has used parametric probability curves to estimate project durations within a Monte Carlo simulation framework.

Another challenge is selecting the correct distribution curve. When we talk about Gaussian, or Normal distributions, perhaps in the context of ’low’, ‘medium’ and ‘high’ three point estimates, we are indirectly intending to use a type of RCF, albeit one that has most likely been fitted incorrectly. The Poisson and Skew distributions often fit the data better, and exhibit the ‘fat tail’ - the so-called rare events whose impacts are highly disruptive to any budget, and which occur more often than people realise!

Regardless of the detailed mathematics, the key point is that we recognise our personal limitations in planning, and that we can overcome these by carefully using much larger data sets that go beyond the knowledge any one person can expect to accumulate in their lifetimes, to make predictions.

The world is changing, and good quality data is becoming easier to access. We can expect greater use of RCF in future.

Support with RCF and Estimating

To find out more about how Project Science can help you with Reference Class Forecasting, please get in touch.