DATE: Fri 24 Aug 2012
BY: Stephen Aldridge
If you own a car, you probably know that after it’s three years old you have to have an MOT every year (if not, you’d better get down to your MOT test centre). The MOT is a basic test that checks that the car is roadworthy – it could still break down on the way home from the test centre, but at least the wheels or exhaust won’t fall off.
I had a client call a couple of weeks ago who had a model they needed to send to a bank, but someone had found an error in it and everyone had got a bit nervous about whether the model had any other errors. After signing an NDA and running some diagnostics on the model, I estimated it would take a good two weeks to do a full review. The client didn’t have two weeks. They had been using the model for quite some time but a few people had made changes to meet short term needs, leading to the error that had been found by chance; they asked what we could do in three days.
The answer was to do something akin to an MOT. The model was basically quite well constructed, so we could look for tell-tale signs that something was not as it should be – inconsistent formulae, any hard coded numbers, checking that the balance sheet balanced (it didn’t but we were able to identify why) and that the cash reconciled. We used our in-house software, and an expert modeller to identify where these sorts of problems might exist then spot checked those areas.
We identified a few errors in that time which the client was able to correct before sending the model to the bank, avoiding the embarrassment and lack of confidence that could arise if the bank had found the errors.
It’s important to realise that this kind of testing isn’t exhaustive – there could still be errors in parts of the model that wasn’t checked. If you want full confidence in a model, there is no substitute for thorough testing and review, which takes a significant amount of time and obviously will cost some money. This cost is likely to be less than either the financial and or reputational impact of the errors that might be found. What this real world scenario demonstrates is that errors do exist in models and that you can manage and reduce the risk of error by testing. How much testing should be done is a question for whoever bears the risk of the errors.
Stephen is a Chartered Management Accountant and has over ten years of financial modelling experience both at KPMG and Deloitte. His early career included engineering, sales and corporate management roles. In 2004, Stephen joined Numeritas as a co-owner and a Managing Director.
We love to hear what you think. please note that comments are moderated so there might be a slight delay. Your email address will not be published.