
THE MODEL FOR PREDICTING TRENDS
Why existing mathematical models fail to accurately understand the future, and how digital ethnography can fix the problem.
The ability to predict how markets will perform and progress over time has become an increasingly sought after skill set among both business leaders and politicians alike.
It makes sense. It can be the difference between a company that grows market share with changing consumer needs, or a company that struggles to keep up. It can be the difference between a candidate winning the White House, or sinking their career.
Yet, despite the weight of responsibility on predictive capabilities, the research and analytics companies that offer such services are subject to very little scrutiny. Most of them employ mathematical models (from basic regression models to Monte Carlo simulations) where outcomes are predicted based on a set of variables. These variables are largely gathered from a combination of rational self-reported data (what consumers say when asked) or consumer behavior tracking data (what they buy, where, how much they spend etc.).
Mathematically, and logically speaking, these all make sense. Especially when one is say, predicting what the size of a line-up will be at a local restaurant on a Monday, or anticipating quality issues on a production line. These situations can benefit greatly from traditional methods of prediction, because the outcome is not being influenced by things like human emotion, societal pressures, cultural context, people’s sense of self, etc..
And herein lies the limitation of these methods.
They miss the underlying reality of human life, and consumer culture. They completely ignore the illogical and hard to explain factors that influence us, as people — factors that shape our realities, influence what we think is cool or important or popular, and what’s not, in a given marketplace, in a particular period of time.
Without emotion, these models lack accuracy.
One need not go much beyond the last election to point out the fundamental flaws in these modelling techniques. Huffington Post’s model showed that Hillary Clinton had a 99% chance of winning the election. The New York Times model showed similar results. As did the models within Clinton’s own campaign team. None of them were right because all of them took a series of rational and logical variables as inputs for their analyses, and ignored the most important factor in predicting shifts in consumer culture — changing consumer beliefs.
Think of a consumer belief as the lens through which a group of people see the world. In any category or marketplace, there are often multiple such lenses (typically 3 to 6) that influence people’s perceptions of the products in the category, and the role they want these products to play in defining their own sense of self.
A similar narrative is expressed in the book Superforecasting: The art and science of prediction, where the authors argue that some of the most successful forecasters are actually people who make judgment calls on the future, rather than use a bunch of different mathematical simulations. They systematically outperform any form of modelling, when tested over the course of a year in hundreds of high stake predictions like the future of oil prices, or the next wave of nuclear warfare proliferation in North Korea etc.. While the book does not aim to identify some sort of a standardized technique by studying these so-called “superforecasters”, for the past 48 months, we’ve been running a series of experiments to do exactly that.
For the last 4 years, our team has been running numerous experiments in prediction, to understand and make sense of these illogical, cultural factors that are driven by the changing beliefs of people in a marketplace. We’ve studied 8 different market categories in the United States, and looked at over 64,000 consumers in the process. Here’s what we’ve learned.
F = Vf
(F)uture = (V)ariables * (f) Prediction Factor
As explained earlier, most mathematical predictive models focus on determining outcomes based on a set of rational variables — most of which are sales/macro-economics/behavior related. They all forget the most important component in calculating the future: A Prediction Factor (f) that is really a function of the emerging set of beliefs shaping a marketplace. Which means, accurately identifying and quantifying these emerging beliefs is critical to the prediction process.
Let’s look at a real-world example to illustrate this.
In late 2014, our team examined the political landscape in the United States, and identified “fairness” as a key Prediction Factor (f) in the 2016 election. That is, through a digital ethnographic study of 8500 voters in the United States, we noted that the candidate that will own and stand for “fairness” will have the best chance of winning the election.
Picking the right variables.
Our world is getting increasingly complex with time. And there’s certainly no lack of variables one can use in predicting future trends in markets. The problem however is that time and time again, the failure of mathematical models have shown us that more variables does not equate to better or more accurate outcomes. In fact, in all the experiments our team ran over 4 years, we learned that just as important as it is to identify the Prediction Factor(s) that matter to a marketplace, so is the task of identifying the variables that are most impacted by these Prediction Factor(s). For example, in late 2014, we knew that the “super delegate” count (in the democrats’ primary process) really did not matter in predicting the outcome of the 2016 election. Yes, it would impact which candidate stood against the Republican on the ballot. But it would not change the likelihood for a candidate to win. Why? Because the “super delegate” count did not relate in any way to the idea of “fairness”. If anything, it negated it. So, just because someone leads the delegate count does not mean they are more likely to win the presidency. The variables that really mattered were things like — how important the issue of “fairness” and “anti-collusion” were to each of the candidates’ platforms, how often they each talked about “fairness”, and how that impacted their popularity on social media, and more importantly, in their rallies. Another key variable was individual donations, and the relationship it bore to the discussion of “fairness”. Here, our team took great interest in the fact that for some of the candidates, every time they talked about “fairness”, they’d see a direct rise in the individual voter (small) donations collected. Just imagine what the Democrats could have done, if they had understood this issue back in 2015.
Estimating outcomes.
In late 2014, with very little information on who was really going to be running, it was impossible to say with clarity who would become the next President of the United States. So, our estimate, calculated using the F = Vf formula stopped at pinpointing “fairness” and “anti-establishment” sentiments as key ingredients for success. In addition, our analysis was also able to identify “collusion between government and corporations”, and the “unfair treatment of patients” as hot button issues that closely tied in with the theme of “fairness”.
Armed with this information, and the key variables our team needed to watch over the coming months, we headed into the primaries with an expectation of revising our estimates as new information presented itself.
As the primaries began, new data was introduced into the equation — the names of the candidates and their past records, outcomes of the debates, attendance at their rallies, individual donations and so on. Tracking these variables, our team was able to identify Donald Trump and Bernie Sanders as two candidates who not only talked the most about “fairness”, but also as the two candidates that garnered an increasing groundswell of support from voters, as evidenced through turnouts at rallies and individual donations received. So, as the primaries went on our estimates began to change. We could see two possible outcomes in the general election.
Here, our formula looked something like this.
Future (F) = Bernie or Trump.
Variables (V) = Turnout at rallies + results of primaries (votes not delegates at key battleground states) + volume of conversation about candidates + focal point issues discussed + individual donations received.
Prediction Factor (f) = The association of the candidate’s brand with fair play and anti-establishment sentiment.
And of course, once the primaries were completed and we knew Bernie Sanders was no longer in contention, our model’s outcome became clear — Donald Trump was going to become the next president of the United States, barring of course, some sort of catastrophic event (like death or illness or the candidate pulling out of the race).
Understanding your industry and the Factors (f) that will shape its future.
The prediction method and model illustrated in the politics example above can be (and has been successfully) replicated in an industry setting, to predict the emergence of new trends that will shape and change the course of your company and the categories in operates in.
Of course, the first step in predicting the future of your marketplace is to identify the Prediction Factor(s) that will matter to your business. There’s only one way to achieve this — through the use of ethnographic research techniques to identify the emerging and changing belief systems that are shaping an increasing number of consumers in your industry. We use the diffusion of innovation model, and apply it to a large scale ethnographic analysis of over 8000 consumers to achieve this in our work with clients. You can learn more about this model here.
Which variables really matter?
Once you’ve identified the Prediction Factors, you can then begin to look at the various variables at your disposal to determine which ones really matter. As highlighted earlier, more variables does not equate to better accuracy. So it becomes critically important to understand which variables are influenced and impacted by the Prediction Factors identified.
Arriving at outcomes.
Once you’ve identified the variables that matter, you then need to gather as much information as possible on these variables and look for relationships between these variables and the Predictive Factors. For example, through the primary process, we noticed that the more Bernie Sanders talked about fighting “Wall Street greed” or “big pharma”, the larger the turnout became at his rallies. The more such relationships you see between variables and factors, the stronger the indications become of a particular outcome over others. But, at the end of the day, arriving at outcomes is an evolutionary and qualitative process. It’s a lot of manual work, and it does not make use of automated models and simulations. Instead, it’s a process that tells us where to look, and what to look for, so we can make estimates that are far better at predicting outcomes than any mathematical simulation can possibly be. It’s a process that looks for compelling pieces of evidence in the form of relationships between variables and their factors, so we have no choice but to arrive at a certain set of conclusions.
Conclusion: No such thing as a perfect score.
Anyone who claims they can accurately predict the future outcomes of markets through an analytical model is lying to themselves, and to you. It’s impossible, mostly due to how our world is increasingly becoming more and more complex. There are not just more variables. They are changing at a pace never before imagined. Will we one day have an algorithm that will be able to keep up with the blinding speed of consumer and popular culture? Perhaps. But the key to success today, is to understand and accept the limitations, to find the best possible path forward.
What our experiments have taught us is that our ability to predict outcomes gets significantly more accurate when we are able to identify and make sense of the Prediction Factors (f) that matter to our sectors, and pin point the variables that are affected by these Prediction Factors. This is the key to implementing a successful trend tracking engine within your organization and making investments at the right time in the lifecycle of a marketplace.