Lessons from the floor

I was approached recently by a friend who, like so many at the moment, had been watching the stock markets. Historic data is available from lots of sources (you can download historic market data into a tab delimited file from http://www.google.com/finance for free), so testing his ‘easy money’ spreadsheet model retrospectively was easy.

It turns out, against all my expectations, that his simple rules-based model did indeed pick up emerging trends, and, more importantly, detect when the trend was about to end and the positions should be cut.

He had dollar signs flashing in his eye, but me, as the rational analyst I am, was not satisfied (encouraging as the signs seemed).  The retrospective testing had relatively few data points, and so the risk that the model was over fitted to the data was high.

So, the next phase was to test it with small stakes in the real world.

Off he went, with his10p-per-point spread betting account to try to prove the model and make his fortune.  A week later we caught up to see how things were going, only to find the account had been wiped out.  Dejected, his conclusion was that he’d fallen into the ‘easy money’ trap and his model was not worth the paper it was written on.

Again, in steps the rational analyst to find out what went wrong…  It turned out that the model never had a chance to prove itself, he hadn’t been betting according to the model, rather to intra-day trends that we thought were emerging from his charts.  He had constructed a model based on data, but allowed his intuition to overrule it.

So now we get to the moral of the story.  Humans are hard wired to see patterns (an elephant in the clouds, Jesus in a slice of bread, Carlos Santana in a Greg’s Steak Slice – Pareidolia is the technical term), and so aren’t to be trusted to objectively analyse data in a graphical form.  Data visualisations help people to identify trends in all kinds of data, from trending topics in Twitter to stock prices, but beware of seeing trends where none exist.  In the case of financial spread betting the cost was the loss of the balance in the account.  For you it could be wasting marketing budget, or setting the wrong prices for a product because you’ve made a decision based on a trend that wasn’t there.

Our MI solutions always combine visual representations of the data with level-headed statistical testing to indicate when something significant has happened.  This enables you to simultaneously keep an eye on the direction of your business graphically whilst automating alerts to tell you when something has changed.

I’m sure you’re wondering how our little story ends now that all of the intra-day charts have been closed and positions are opened or closed based on the output from the model…  Watch this space!