Addiction to Prediction
January 3, 2012 § 11 Comments
With each year end, all forms of media spew a tidal wave of predictions. From the apocalyptic to the mundane, we get predictions from prognosticators on who will win an Oscar to which Republican will win in Iowa to how well the market will do in 2012 to who will win the Super Bowl. But it’s not only during year ends that we get a hefty dose of soothsaying. It’s a public non-stop obsession. Dare I say, it’s an addiction. Predictions are in every facet of society – within industry, we are constantly trying to get insight on the level of demand this month for our products; the level of prices within each product type; and which company will gobble up which other company. The fact that foresight can be a key advantage when competing for resources and competitive superiority is not surprising. What is surprising is the amount of noise pollution and the insatiable desire to listen to that noise.
What is an Expert?
I still hear my favorite finance professor lecturing during one of my b-school classes about “experts”. He illustrated quite powerfully (obviously, it’s stayed with me for all these years), how poor predictions were made by economists on interest rates, GDP growth, oil prices, and stock prices among many other measures. In article after article, economists, industry experts, political experts, scientific experts were shown to be just slightly better than random guessing. What’s worse, most “experts” tended to influence each other, so that consensus predictions prevailed. The group of economists’ predicting the direction of interest rates tended to lump together in narrow ranges which indicated that working from the same sets of data with the same sets of assumptions, they also tended to create the same range of estimates.
Risk and Probability
We all know that the future is uncertain and that many unknown factors impact future events, yet we go to great lengths to predict. The bottom line is that we can draw conclusions that are more about probability than actual pinpoint calculation. If we normalize probabilistic outcomes for earnings per share for Apple this coming quarter, we can estimate EPS outcomes within ranges. If we believe published consensus estimates by analysts, we can find that mean estimates are at 9.81 with a coefficient variance of 4.39. Statistically, this variance is only significant for historical relevance and should not be seen as a predictor, but given analysts do not have crystal balls, they still use it as the main factor for setting probabilities. So, if we conclude there is a 95% chance that EPS will fall within the range 8.56 – 11.06 (or two standard deviations), we are essentially placing bets based on probability. Now, the valuation of a share of Apple common stock will vary greatly depending on where in this range actual EPS falls. Of course, there is still the 5% chance that EPS falls outside the expected range. And further, these numbers are purely estimates based on one set of assumptions that no two analysts would ever agree on.
When looking at all these predictions, it can quickly become apparent which “experts” are really viewing their data through a critical lens and which are simply along for the ride by echoing other expert’s viewpoints. What I find most discouraging is how confident some prognosticators are, especially those on television and web broadcasts. They emphatically proclaim their view in an effort to persuade viewers they are right – trying to create self fulfilling prophesies through persuasion – perhaps the most egregious offense. We see this regularly on political discussion panels where party-aligned or candidate-partial analysts make their case for persuading us what people really want and how they will vote. Are they really giving us a scientifically sound viewpoint or simply trying to manipulate our view about what will be?
Predicting Human Behavior
The digital age has provided a powerful platform for gathering information on individuals’ behavior. Companies can gain insight into buying behaviors as well as data on individual and group interests in entertainment, politics, as well as professional and social connections. The usefulness of this information is at once obvious and complex. For instance, if we know that there is better than 50% chance that a person buying a camera will also buy a camera case then it would be an effective sales practice to suggest a camera case at the point the buyer selects a camera. This practice is quite common now with online purchasing. We also see it fairly frequently with phone sales as well as with fast food ordering; e.g.: “Would you like fries with that?”. Recently, I was shopping for a new lawn mower and researching options on the website homedepot.com. Within several minutes, after performing a search on the site, a pop-up offer to chat with a representative came up. I took that offer as I had some questions about the models. After about five minutes, the rep offered me a 10% discount if I’d like to purchase the item online and he’d help me through the purchase process. I had already made the decision to buy the item, so I was happy to get this discount, but I wanted to pick it up at a store near me rather than have it shipped. Their process allowed for this flexibility as I could purchase online and the item would be instantly put aside for me to pick up that evening. This process was ingenious. What percentage of people shop to the point of sale and then drop, reducing the chance of buying the item through the original site or perhaps not buying that item at all? This new discount offer through the chat rep can help homedepot.com reduce that drop percentage. Now, Home Depot does not know if I would have purchased that item that day online anyway and they do not know if I would have gone to the store and been willing to pay full price. The fact is: there will be some percentage of margin they relinquish for the sake of capturing a higher percentage of potential sales. Home Depot, like so many other retailers, banks, insurance companies, and drug companies are in the business of prediction – predicting what you will do if they communicate with you in a certain way at a certain time.
Statistical Sampling – A Foundation for Predictions
Statisticians often tell a joke about a man with his head in a refrigerator and his feet in an oven – on average he feels about right. Sampling is used to determine probabilities and to make decisions on the level of risk we are taking. It’s sampling and probability that determines the rate we pay for life insurance, car insurance, and all other types of insurance. So, when we try to predict Apple’s earnings per share for next quarter or whether a catastrophic disaster will strike an Asia Pacific country next year, we must calculate the odds, the percentages, the probabilities. The fact remains, however, that not all outcome distributions fall into a normalized curve and highly unlikely events can be game changers.
Prediction should be all about risk, uncertainty, and likelihood, but what you’ll hear this week and throughout the year is a chorus of experts telling you with great certainty what the future will bring. Don’t believe them. If you’re jonesing for advice, try listening to those who are providing detail on probability, risk and trends. But know that the future is never about certainty and always about probability. When prognosticators get it right, they were just plain lucky. They may have played the odds. They may have had some truly intuitive insight that others did not. But there is never a sure thing.