The Making Or Breaking Of A Forecast - Predictive Analytics
In my very beginnings in the finance industry, I had the chance to do an internship in the Treasury department of a big Austrian credit institution. In those days, the department had organised a regular competition among the staff members. It was all about forecasting target prices of traded assets for the upcoming weeks... be it currencies, options, shares... you name it! After each week passing, they would gather to see who best foresaw the market, i.e. who had the best expert domain and feeling for the market.
It was a cute contest. Albeit, if we put things in the context of a data driven world we are living in now, this is exactly how forecast should not be done as it has low chances to succeed. Setting a point estimate (like a target price) based on some personal overview and/ or opinion about the market development and see how close you can get is something we might have done in the past, but now is becoming obsolete. Sure - sometimes you will score, you might even score several times on a row, but inevitably, other times you will be off-market and get tricked by your subjective opinion and perspective.
Why? Because doing forecasts this way is nothing else but 'educated guessing'. It is the look into the crystal ball trying to get some answers. In the longterm it will not work, even if you have some hits in-between. Your business success depending on some old methods is at least a lost opportunity.
On the other hand, data driven predictive analytics taken seriously is all about not moving in any of those directions. There are no crystal balls to look into for the purpose of predicting the big future. Though, there is a lot of data ready to be analysed and ready to unveil patterns in order to get a grip on how a market or a certain market position could objectively develop in the future.
Predictive analytics starts with the right questions to ask within a certain business context. Needless to say that in this context the availability of a lot of business experience is necessary. Collecting, cleaning and processing data allows us to do the first step, so called "prescriptive analytics" and to see where the journey could go.
Subsequently, the appropriate predictive analytics method is to be chosen depending on the data and the pending questions in place. There are a lot of highly effective predictive analytics approaches out there. Their effectiveness and accuracy depend on the circumstances in place. The scope goes from relatively simple Monte Carlo simulations, to Machine Learning models, Deep Learning applications or Networks Dynamics fabric. And yes, sometimes there are statistical models to be incorporated. In fact, in a lot of cases there is a significant amount of statistics running in the backyard.
In a nutshell, predictive analytics helps you in a data driven and structured way to navigate through a messy and noisy world. It is a tool which helps to understand how exposed a certain position is due to expected developments to-be. The crucial thing is, you have to ask the right questions resp. you have to know the industry you are analysing.
Once you have asked the right questions and did your analytics part it is quite important to learn how to use the results and to understand that they are always coming in terms of probabilities. You have to learn to use them and read them, as it is the case with any other analytics tools. As there are no free lunches, there are no sure bets! Treating predictive analytics like a crystal ball is for sure not doing anybody any favours.
Certain developments are more likely than other developments. This is what you get. And this term 'more likely' is not expressed as a point estimate. It is expressed as an area of possible developments.
Let's go back to the real estate business for an example: A statement like "I expect the rent price to hit EURO 25.5 by the end of the year" is nothing more than an educated guess and is therefore useless. But you could state instead that a rent price rising above EURO 25.5 within the next period is significant more likely than a rent price rising just above EURO 24.0. So when you learn to read the results it will offer you the same answer you were searching for but not as an instant solution but as a result of the industry knowledge combined with tech tools.
This "forecast" you get in a structured, reliable and transparent way. Provided this, a decision maker is capable of re-evaluating e.g. a position of a portfolio in a current market or to decide on the timing of the entrance into a new market in a very comprehensible way which could be extremely helpful when a decision already made is scrutinised years after.
Once again, industry knowledge is crucial! And here I mean on both sides. It is very helpful when the tech team has a profound knowledge of the industry to know which questions to ask and which tools to use. But even more important is that both sides are involved in the interpretation of the results and deciding on which of the highlighted scenarios has the biggest potential to succeed. It is a new way of decision making, data driven decision making and we have to learn how to do it!
Why is this important?
Basic goal of predictive analytics is to distill the relevant signal from all the noise which is out there in the market. It is therefore a tool to get a certain orientation in a messy market environment, hence to reduce uncertainty.
Predictive analytics helps a decision maker to constantly evaluate the risks or opportunities the company is exposed to. On the basis of this constant and structured evaluation, companies are able to act in an early stage in contrast to just reacting to already unfolding events. And this makes the company a decisive step faster than the competition.
This is a good way also to countercheck decisions made by management boards in turbulent times. Recently I made it through financial reports of several real estate companies and realised that some of them made write offs and write ups in the same year, without clear business indicators.
The fact is that this almost erratic behaviour gives a very good impression of the inherent uncertainty prevailing in the current market circumstances with all its risk implications. The questios if indicators used are really reliable is something shareholders and investors are for sure interested in. This is a classic application case for predictive analytics.
And why not to use it? Sometimes you should render any help you can get.