What is the correct assumption? Or how to choose a good set of baseline assumptions
In a meeting to review business case projections, the discussion ranged over various assumptions in the financial model and someone asked “But what is the correct assumption?” For those with a background in analysis, this question was an amusing distraction, but it raises an important point – what assumptions should you use in your business forecast?
One of the problems of any forecast is that it reflects the biases of the forecaster. And when you build a financial model of your business you are also making potentially biased structural assumptions about key business drivers. So, even though you’ve constructed a ‘rational’ financial model, it will inevitably be biased and the assumptions potentially unrealistic or inconsistent.
So what should you do to minimise bias? The best approach is to view all assumptions as just that, assumptions, and not facts, and to use sensitivity analysis to refine your understanding.
Sensitivity analysis helps answer the “what happens if…?”questions, such as sales X% down on forecast, payroll inflation up by Y%, or overheads reduced by Z%. A good financial model will have the capability to do this quickly and easily, perhaps with switches for key variables or scenarios.
The main benefits of sensitivity analysis are to evaluate the range of potential outcomes, to understand better the relationships between inputs (assumptions) and outputs; and to test the robustness of the model by revealing potential errors (as highlighted by unexpected, counter-intuitive or non-linear relationships between variables).
If you use sensitivity analysis systematically by adjusting your assumptions incrementally away from a central value, rerunning the model and reviewing the output, you can start to understand which variables are the most important to your business. A more sophisticated version is to use the Monte Carlo method to generate a large number of outputs having randomised each of the key input variables. Ideally you will automate this process and then run the model through a large number of iterations.
However accomplished, reviewing the range of outputs will give you a sense of the most likely outcome, and the potential range around that expected outcome. This process also provides a ‘sense check’ for your baseline assumptions.
There is another useful variant of sensitivity analysis to consider before finalising your baseline assumptions: break-even analysis, which allows you to find the levels for key variables where the model breaks even, for example sales volume, selling price or cost of sales (whilst holding all the other variables at their central assumption). Break-even analysis is a simple and powerful concept, but surprisingly under-used.
So, having reviewed your sensitivities and evaluated the break-even points, you have arrived at a better understanding of your assumptions. You may find that something previously considered insignificant is a critical, or the impact of some variables is reduced by self-correcting mechanisms, others may affect timings but not quanta, and yet others will be much less significant than the received wisdom would have suggested. You should also consider each assumption in the light of the others: are you assumptions mutually consistent? Such insights may provoke further research and debate as you home in on your baseline assumptions.
You will probably have realised by now that there is no such thing as a “correct assumption”. The best you can have is a baseline set of realistic assumptions that are internally consistent. What you do need is a management process: a progressive approach to a better understanding of your assumptions leading to a better forecast of the outcome.
This review process should continue every time you update the model with actual performance. After an update that results in a step change from the original plan, ask yourself “Is this the start of a trend, a blip, or a timing issue?” Keep running sensitivities (especially those suggested by recent actual performance) and, if necessary, update your assumptions. You should converge on an increasingly accurate forecast which, after all, is the goal of your modelling.