Predictions are not hard once you know the outcome...

“Man is a deterministic device thrown into a probabilistic Universe.” Amos Tversky (cited by Michael Lewis)

My first reaction after reading “The four global forces breaking all the trends” by Dobbs et al., was to picture myself conducting a hike in St. John’s, Newfoundland.

I can make a few predictions with great accuracy about that hypothetical hike. Unfortunately, there are items of importance that I can’t predict at all. Moreover, the big initial hurdle to my forecasting exercise is that it is difficult to figure out which ones are which.

Let’s look at some of them: 

- Will it rain?

- Will my plain ticket cost me exactly US $378.99?

- Will my airline of choice be a Canadian or American one?

- Will my ride to downtown be more than 15 min.?

- Will I be able to watch a whale if I do the North Head Trail? What about the Quidi Vidi Trail?

For instance, the odds of watching a whale while walking the Quidi Vidi Trail? Easy to predict or hard? Easy! I feel that I can predict with 100% accuracy that it won’t happen. However, I have no idea what my real odds are if I choose the North Head Trail, instead (http://www.newfoundlandlabrador.com/things-to-do/hiking-and-walking). I feel that is also easy to predict with a 100% accuracy that my airline of choice will be Canadian, since I’m flying from NJ, but I have no idea what the price of the ticket will ultimately turn out to be. Apparently, neither do most people, according to the authors of “Airfare Prices Prediction Using Machine Learning Techniques” (https://www.eurasip.org/Proceedings/Eusipco/Eusipco2017/papers/1570348051.pdf). Will it rain? I feel confident, although not to a 100% level, that it will. 

We are intuitive forecasters and we successfully exert our “visionary” powers on events that are simple, are broken down in small pieces, and happen all the time. Since the Quidi Vidi Trail goes around a lake, I can infer with great accuracy that there will be no whales coming into my field of vision during my hike. Analogously, since no American airline flies into St. John’s, I can predict that I will fly a Canadian one when I visit there (the odds go down significantly if I’m coming from Europe). Last but not least, the 212 days of rain that have been recorded by the weather authorities in Canada as the yearly average for the city make me feel confident that I will need rainwear.

Moreover, once events take place, common sense leads us to believe that we can put together the reasons of why they happened. If I finally fly to St. John’s and don’t see a whale during my North Head Trail hike? It must be because I went there during the Summer. As easy as two plus two! We are lousy at predicting complexity but we are excellent at explaining anything and everything. 

Everybody has a theory on why Harry Potter, Google and iTunes came to be great successes. But Rowling’s book was rejected by 12 publishers, Google was for sale in the 90’s and had no takers, and iTunes and the iPod replicated the same strategic “mistakes” Sony made with their Betamax product and ended up reaping very different results. It’s easy to get an explanation of what were the factors that mattered for an outcome to emerge, once we know the outcome and we can piece together a sequence of events that seem to have mattered.

The proof that most, if not all, of those explanations are wrong is that nobody has been able to monetize their model of “what made the outcome certain” and replicate the next iTunes. Or Google. Or Harry Potter. We have “learned” to tell their stories but we still can’t predict with them.

Where the authors of the paper say that the four forces will compel executives, policy makers, and individuals “to scrutinize their intuitions from first principles and boldly reset them if necessary”, I feel that they are missing the boat. It is not that there is a need to set intuitions differently. I think that we need to abandon them as tools to predict what is essentially unpredictable. What is the basis for the statement? The findings of researchers who work on the field.

Philip Tetlock, a psychologist, has been investigating forecasts and professional forecasters for over twenty years. One of his tests, focusing on complex societal events lasted that long. When he reviewed the results, on hundreds of these so-called experts, they were not very encouraging. The experts performed only slightly better than random guessing.[1]

Corporate strategies, the effects of government policies and societal trends are not simple and cannot be broken down in manageable pieces where cause-and-effect is contained and easy to see. Asian cities may be where the markets are moving but if the average temperature of the planet rises by 3 degrees C, the Ocean will submerge most of them. Want to place a bet on what will be the outcome?

The best we can do when forecasting complex events is to be humble, look at the plausible different scenarios as probabilities, use data and trends where applicable and admit that, oftentimes, we don’t know what we don’t know. I believe that just knowing the true limits of what we can predict and abandoning the temptation of using our intuitions will make strategy development more realistic and useful, although not necessarily as accurate as most would like it.

In any case, pack an umbrella when traveling to St. John’s!

Leoncio Estevez-Reyes, lwe@andrew.cmu.edu

---

[1]Review of: Philip E. Tetlock. 2005. Expert Political Judgment: How Good is it? How Can We Know? By Adrian E. Tschoegl and J. Scott Armstrong UPenn. June 2008. SSRN Electronic Journal. DOI: 10.2139/ssrn.1153126