What Can We Predict?
This review does not encourage me to read Dan Gardner's book "Future Babble". Gardner concludes that even the experts have trouble predicting the future. Shocking! What may be more surprising is that we have had great success in predicting the future.
A "small ball" example is the Zagat's restaurant guides. The quality of a meal at a restaurant you have not been to is a random variable. To help you make a "better choice", millions of people have studied the Zagat's guide before they invest and walk into a restaurant. If people feel that the book gives bad advice, then they would never refer to it again and they would tell their friends and Zagat's would collapse as a business venture. Due to competitive market forces, Zagat's has an incentive to accurately aggregate popular opinion (and it does this by averaging respondent's rankings). The proof that Zagats can predict the future is that people keep buying their books. Apparently, a restaurant's past success predicts its future success.
Another example: predicting Presidential election results. Ray Fair has demonstrated that a very simple statistical model can predict such election results. Here are the data. You can play around with that.
So, the author is correct that Nostradamus was a special dude but we continue to have examples of our ability to predict.
Why are some websites that help you profile who should go out on a date with successful? Match.com immediately explains how they engage in profiling using the information you provide to predict who might be a good match for you. They have the right incentives to have a good predictive model ---- otherwise nobody would sign up for their services.
If everything in life is "random" , why do firms seeking to hire employees interview people? Why don't they randomly select a resume from a stack and just hire that person? They don't do this because they learn things about you from the interview and these insights are predictive for whether you a good match for that firm.
So, "macro prediction" is hard --- but "small ball" micro prediction has a bright future.
I am a little bit defensive here because suppose that it is true that future events are uncorrelated with past events and suppose that it is true that climate change is raising the probability of horrible "fat tail" events, then we can be struck unexpectedly with shocks that we never anticipated and thus will be unprepared for them and will suffer greatly.
To begin to be optimistic at all about adapting to climate change, we need our entrepreneurs to be able to anticipate broad future trends (i.e to be able to predict likely scenarios under climate change). We need the population to anticipate increased risk of living in certain locations (i.e coastal areas at risk to flood). If flood locations and severity is completely unpredictable then even people who want to protect themselves will have no idea where to locate themselves to protect themselves. Now, climate scientists wouldn't bother to study long run data patterns if it offered no clues about the future.
A "small ball" example is the Zagat's restaurant guides. The quality of a meal at a restaurant you have not been to is a random variable. To help you make a "better choice", millions of people have studied the Zagat's guide before they invest and walk into a restaurant. If people feel that the book gives bad advice, then they would never refer to it again and they would tell their friends and Zagat's would collapse as a business venture. Due to competitive market forces, Zagat's has an incentive to accurately aggregate popular opinion (and it does this by averaging respondent's rankings). The proof that Zagats can predict the future is that people keep buying their books. Apparently, a restaurant's past success predicts its future success.
Another example: predicting Presidential election results. Ray Fair has demonstrated that a very simple statistical model can predict such election results. Here are the data. You can play around with that.
So, the author is correct that Nostradamus was a special dude but we continue to have examples of our ability to predict.
Why are some websites that help you profile who should go out on a date with successful? Match.com immediately explains how they engage in profiling using the information you provide to predict who might be a good match for you. They have the right incentives to have a good predictive model ---- otherwise nobody would sign up for their services.
If everything in life is "random" , why do firms seeking to hire employees interview people? Why don't they randomly select a resume from a stack and just hire that person? They don't do this because they learn things about you from the interview and these insights are predictive for whether you a good match for that firm.
So, "macro prediction" is hard --- but "small ball" micro prediction has a bright future.
I am a little bit defensive here because suppose that it is true that future events are uncorrelated with past events and suppose that it is true that climate change is raising the probability of horrible "fat tail" events, then we can be struck unexpectedly with shocks that we never anticipated and thus will be unprepared for them and will suffer greatly.
To begin to be optimistic at all about adapting to climate change, we need our entrepreneurs to be able to anticipate broad future trends (i.e to be able to predict likely scenarios under climate change). We need the population to anticipate increased risk of living in certain locations (i.e coastal areas at risk to flood). If flood locations and severity is completely unpredictable then even people who want to protect themselves will have no idea where to locate themselves to protect themselves. Now, climate scientists wouldn't bother to study long run data patterns if it offered no clues about the future.
posted by Matthew E. Kahn at 12:26 PM
0 Comments:
Post a Comment
<< Home