“Prediction is very difficult, especially if it’s about the future.” - Niels Bohr
why prediction science?
what is prediction science?
Humans have been trying to predict the future at least since the beginning of recorded history, and probably ever since we first comprehended such natural regularities as day following night and the moon passing through phases of illumination. Prediction is ancient; prediction science is a fairly young discipline.
What makes a prediction scientific? The essence of science lies in making our beliefs about the world testable through observation. Most of us learn the fundamental principles of the scientific method by the time we reach high school:
Observations of the world lead to hypotheses or possible explanations for what we see.
Experiments are conducted in which new evidence is gathered in a way that permits disconfirmation of the hypothesis or explanation.
Based on the new evidence, hypotheses may be abandoned or revised, and the cycle repeats itself.
Prediction science is not a single discipline; statistics, mathematics and computer science are important elements. Prediction science also draws ideas and methods from behavioral science, economics, and biology.
We humans appear to be uniquely equipped to make naïve predictions about the future. We have what appears to be an innate ability to detect patterns and to infer causal relationships. These tendencies are so powerful that we often believe we’ve discovered meaningful patterns and causal relationships in random data. Consider the belief in the “hot hand” performance streaks that some athletes and gamblers seem to experience occasionally. Statistical analysis of actual streaks shows that they are indeed random, but belief in the hot hand persists.
We are also susceptible to explanations and predictions that “make sense” whether or not they have any grounding in reality. Our susceptibility to a good story can lull us into a false sense of confidence in our ability to foretell the future. When it comes to making “bet the business” decisions, this can easily lead to disaster. Even when we try to be systematic and use evidence-based tools like data mining and predictive modeling, experience has shown that we often do only slightly better than random guesses.
These predictions—whether intuitive or evidence-based, most often fail for one of these five reasons:
Our underlying causal model of the market is either too simple or too complex.
The prediction relies solely on statistical associations in historical data with no accompanying causal model.
The prediction ignores the baseline or expected probabilities of key events (the “known knowns”).
The prediction relies on untested assumptions in place of real evidence for the “known unknowns.”
The prediction does not incorporate the possibility of “unknown unknowns”—cause and effect relationships that we have not yet envisioned or discovered.
Our prediction science solutions specifically address these five factors and offer a countermeasure to the unscientific and naïve ways we try to imagine and anticipate the future.
prediction science tools
Predictive modeling is an important tool for testing hypotheses around our underlying causal model of the market. Using historical data we look for patterns that can be turned into testable propositions.
Agent-based simulation is a powerful tool for untangling complex interactions in the market place and for discovering possible and probable futures where little or no historical data exists.
Bayesian analysis gives us a way to quickly update our predictions as new information becomes available and to evaluate the likelihood of our hypotheses when we have limited information.
Game theory simulations provide insight into the likely outcomes of competitive interactions, such as response to pricing changes.