What does Econometric Modelling do?

In an ideal, ‘laboratory’ environment we would be able to run exhaustive tests, turning on and off one potential sales driver at a time to see what effect each of them had on sales. In the real world, this doesn’t happen. Typically, all levers must be pulled at once to maximise effectiveness. This makes it hard to know for sure which of your sales drivers are doing the work and delivering results.

That’s where econometrics comes in. By looking back over enough historical data for your media spend and activity – e.g. when you did 40 ratings on TV in the first week of May, when you spent £100,000 on press in the middle of June, and so on – a well-structured econometric model observes the correlations between what you did and what happened to sales as a result. From there it calculates the effect that each activity in turn had on sales.

You can see how powerful this is. It means, as marketers, that you can actually prove that A, B and C marketing activities led to X, Y and Z results.

More than marketing

Of course, it’s not just the marketing activities you did that affected your sales. Seasonality, price, the weather, etc. all played a part, in many cases a bigger part, than your marketing efforts. So, a good econometric model will also include non-media drivers of sales, such as seasonality, price changes and so on.

In fact, we can go further and say that a model that doesn’t have these other, non-marketing drivers is a bad model, and it is going to give you wrong results by overestimating the role marketing activities plays. A bad model can show you in a good light initially – your marketing activities can look really successful – but it will come back to haunt you as you will be making decisions on inaccurate data and committing spend that won’t deliver results.

The USP of econometrics is that it is the only technique capable of measuring the impact of simultaneous drivers, retrospectively.

How does Econometrics work?

Put simply, Econometrics takes a bunch of data (called variables) and works out the relationships between them. It’s a slightly strange beast in that it combines elements of mathematics, statistics and economics. Fortunately, others have gone before us who have done much of the hard work – proving theorems, solving equations – to develop for us what we have today. Others have, more recently, programmed computers to automate the handle-turning part.

It’s all about relationships

Take two series of data that appear to be correlated. Plot them on an X-Y chart. In fact, don’t bother. Just look at the one below.

Now, using your mental ruler, draw a line through the data points that best fit the relationship you can see there. Congratulations! You’ve just built an econometric model and you did it in your head.

If you’re someone who uses the trendline feature in Excel, then you should know that all Excel is doing is building an econometric model (a very basic econometric model, very quickly, in the background) and showing the result of it. In fact, that’s where the equation for the line in Chart 2.2 came from.

This works when we’re looking at the correlation between only 2 data series. What if we were to give you now a set of 20 data series, all of which are, to a greater or lesser degree, correlated with each other?

This is where our mental econometrics software fails us, chiefly because most of us don’t find it easy to visualise charts in 20-dimensional space.

Computers do the heavy lifting

Instead, we need some calculus, some statistics and, above all, a good computer package to do the heavy lifting. But – and here’s the important thing – the calculus, the statistics, and the computer package aren’t doing anything in principle that you didn’t just do a moment ago when you fitted your mental line through all those dots. It just does it for more data than we can manage, and, it must be said, a little more accurately than our heads are capable of.

We are oversimplifying; there is loads more that a stats package does, a lot more efficiently. And, as people who cut their teeth running econometric models on pocket calculators in university exam halls, we’re profoundly grateful for the advances in computing over the last 25 years.

There’s another important point here, all stats packages fundamentally do the same thing. Give them the same data and they will all spit out the same results.

The clever part of building an econometric model is knowing what variables to put in your model, what permutations to test, how to interpret what the stats are telling you and so on.

Conclusion

Econometric Modelling allows Marketers to answer the question: ‘What is driving my sales?” and is the only technique capable of measuring the impact of all simultaneous drivers, retrospectively. It uses mathematics, statistics and economics to identify patterns and correlations that will help you answer that burning question. Used cleverly, a good Econometric Model can help you to make key marketing decisions and shape your future marketing activities to deliver a stronger return on investment.

If you’d like to find out more about Econometric Modelling and whether you should run an Econometric project, download our free Ebook ‘A Guide to Econometric Modelling for the Modern Marketer’ or drop us a line.

Sam Watts