Cruncher is the pseudonym of an actuary working in London with experience in insurance, pensions and investments.
More and more areas of life involve long term financial models. The results of these models can affect thousands or even millions of people and billions of pounds. So it more important than ever that these models are up to the job. For example the UK Government was left with red faces and a very large bill when the model used for granting rail franchises was found to be not good enough.
So what makes a good model? Fundamentally you need an experienced competent person or team running the model. Beyond that there are some basic features that good model should have. I have listed 9 key features but there are plenty of other features that might be desirable for a good model. Equally there may be times when you have to disregard one of these principles. But if you do you should know you are breaking the rules and you should have a very good reason.
1. The model must be able to reflect the risk profile and possible outcomes of the whatever is being modelled.
Remember that if you don't build a feature into a model, then it's no surprise if is doesn't turn up in the model results. If it's important, model it. If it's not important ignore it, and know that you have ignored it. All models are incomplete. Good modellers know where the gaps are and why they don't matter.
2. The model must be valid, rigorous (enough) and clearly documented.
In other words. Make sure the model does what you say it will. Make sure it will do it in any conceivable circumstances. And make you clearly record what you have modelled, how you have modelled it and why.
3. When designing the model consider how you will test it.
If you can't adequately test the model, how can you be sure that it is correct? Ideally test it against a manual calculation or a trusted independent model. This is of course over and above ensuring that your model is theoretically correct (i.e. doing that you think it should).
What are financial models used for?
Financial models are used somewhere in almost any kind of finance product.
Models are used to price insurance products and to make sure insurance companies hold enough money in reserve to pay for claims in the future (and without that your insurance policy would be worthless).
Models are used to price complex investment products where there isn't an obvious market price.
Models are also used by financial planners when they help you plan your finances.
Almost all areas of our financial lives are affected by models so it's important they are done right.
4. Key features should be parameterised and parameters should be set carefully taken into account the purpose of the calculation.
It's good practice to make any key features of the model that might change, such as an interest rate, parameters so that they can be changed without changing the model itself. This is to make life easier for everyone and reduce the need to testing the model in depth - which you would need to, if you changed the model itself.
These parameters will of course affect the result of the modeling so you should make sure that they are appropriate for whatever task you are currently working on. You may need to gather data, or ask key decision makers to justifythe parameters you are using.
5. The model should be a simple as possible and easy to communicate.
Not all models will be simple. It will depend on what you need to model (see point 1). But adding bells and whistles to a model when it won't change the results in any meaningful way only increases the time and costs spent to no gain. It also makes it harder to spot and/or fix errors. And it also means that it will be harder to explain exactly what the results mean.
6. The results should be easy to find, clearly defined and in a format that can be explained easily to the end users.
It is no use having a really clever model if you have to trawl through pages of irrelevant information to find what you want. That increases the dangers that users end up picking out results that are not what they think they are and using them inappropriately. For the same reason, all results should be clearly and unambiguously labelled.
7. Avoid implying that everything can be modeled
Unless your model is the size of the universe, it will have limitations. It's better for you and end users if you know what these limitations are and are up front about them.
8. A range of methods of implementation should be available.
This makes it easier to test and parameterise the model and gives focus to the results obtained (see point 6).
9. You should be able to develop and update the model without too much difficulty.
It's unlikely that any model achieve it's final form immediately. Users may come up with new requirements and circumstances may change meaning that the model needs to be updated. (And this is yet another reason why documentation is so important.)
Finally, a warning. Don't treat that complicated black box with too much reverence. It is just a tool, made by people for people. Nothing more. Models are great tools, but really bad masters. Master your model and make sure it serves your purpose, not the other way around.
Sumit Arora from India on May 02, 2016:
I think Great model should tell how much money should you invest and what will be its return after 10 years. Then, it can be successful.