Do you remember the proofs that you did in high school geometry? I have been thinking about forecasting lately (as a topic of keep professional interest and personal curiousity) and am wondering if I could use this proof concept to make some connections. See if you can follow the following:
1. Each one of us has a bias.
It may be conscious or subconscious. It might be easily seen by others or by ourself. In any case, there isn't a person out there that is free from natural inclinations, pessimism/optomism, strengths / weaknesses, etc.
2. Bias create errors
A optomistic person may see the glass as half-full, and credit people with too much capability than they have demonstrated. A pessimistic person might be overly conversative and play out the "worst case scenario." Reality, in either case, will be different than the bias of the individual. Even a small error is still an error.
3. Bias manifest themself in every decision point.
Unless we actively fight it (often by overcompensating in the opposite direction), we demonstrate this bias at every point in our life where we are asked to make a choice.
4. Forecasting is all about making decisions.
Forecasting is an estimation process in unknown situations, which inherent relies on our ability to extrapolate our beliefs, data, and assumptions and make decisions. Whether forecasting financial results, anticipating the success of a new product, informing supply chain decisions, guessing when the train will arrive, or charting out your personal retirement savings, it is all about guesses, and educating those guesses as best you can.
5. The more decisions, the more error
So, if the above are true, each decision point in a forecasting process would give you an opportunity to illustrate your bias, introducing error. This error would multiply by every decision point to amplify your natural bias.
6. Corrolary: Fewer decisions, less error
Some forecasting processes having fewer decision points, thus would inherently be more immune to bias error as described above.
7. Conclusion: Forecasts based on limited data points are inherently more accurate than those based on multiple decision points.
Sounds logical, right? But think about this. This theory would say that forecasts based on historical trend data (one data stream) are more accurate than bottom's up forecasts which roll up the estimate sales targets from multiple sales territories. This is the counter intuitive part. When asked for a more accurate forecast, what do most general managers or business owners do? They go head first into bottom's up sales data to prove their assumptions, to show where the money will come from, and the like. This seems like it would be highly accurate and give them more confidence than simply extrapolating past performance. But I am coming to the conclusion that it isn't true. More analysis, leads to more error, which leads to more analysis, and forecasts get wildly out-of-whack.
So, what the is the true north for forecasting? I asked this question of several folks whose forecasting ability and experience I respect and they replied, "past performance is the single best indicator of future performance."
This phrase is very familiar to those who are experienced with behavioral interviewing techniques, which seek to illustrate a candidates competency and compatibility for a new job by matching the patterns of performance and styles of work from their past to the ideal for the position. This theory, in human terms, says that as you grow older and gain experience, you simply become a more refined version of the person you always were (a key point in Marcus Buckingham's new book entitled Go Put Your Strengths to Work).
If you are interesting becoming a better forecaster and trend observer, you'll like this post showing Paul Saffo's (Institute for the Future) Rules for Forecasting.
Enough philosophy. How do you forecast? How do you take out the bias, but ground your forecasts in reality? Are complex methods better than simple methods? How important is an understanding of the business (thus the passion that would lead to bias and to errors) or is it better to have a "pattern recognition" person that doesn't know the business? Your insights and experience greatly appreciated.
1. Each one of us has a bias.
It may be conscious or subconscious. It might be easily seen by others or by ourself. In any case, there isn't a person out there that is free from natural inclinations, pessimism/optomism, strengths / weaknesses, etc.
2. Bias create errors
A optomistic person may see the glass as half-full, and credit people with too much capability than they have demonstrated. A pessimistic person might be overly conversative and play out the "worst case scenario." Reality, in either case, will be different than the bias of the individual. Even a small error is still an error.
3. Bias manifest themself in every decision point.
Unless we actively fight it (often by overcompensating in the opposite direction), we demonstrate this bias at every point in our life where we are asked to make a choice.
4. Forecasting is all about making decisions.Forecasting is an estimation process in unknown situations, which inherent relies on our ability to extrapolate our beliefs, data, and assumptions and make decisions. Whether forecasting financial results, anticipating the success of a new product, informing supply chain decisions, guessing when the train will arrive, or charting out your personal retirement savings, it is all about guesses, and educating those guesses as best you can.
5. The more decisions, the more error
So, if the above are true, each decision point in a forecasting process would give you an opportunity to illustrate your bias, introducing error. This error would multiply by every decision point to amplify your natural bias.
6. Corrolary: Fewer decisions, less error
Some forecasting processes having fewer decision points, thus would inherently be more immune to bias error as described above.
7. Conclusion: Forecasts based on limited data points are inherently more accurate than those based on multiple decision points.
Sounds logical, right? But think about this. This theory would say that forecasts based on historical trend data (one data stream) are more accurate than bottom's up forecasts which roll up the estimate sales targets from multiple sales territories. This is the counter intuitive part. When asked for a more accurate forecast, what do most general managers or business owners do? They go head first into bottom's up sales data to prove their assumptions, to show where the money will come from, and the like. This seems like it would be highly accurate and give them more confidence than simply extrapolating past performance. But I am coming to the conclusion that it isn't true. More analysis, leads to more error, which leads to more analysis, and forecasts get wildly out-of-whack.
So, what the is the true north for forecasting? I asked this question of several folks whose forecasting ability and experience I respect and they replied, "past performance is the single best indicator of future performance."
This phrase is very familiar to those who are experienced with behavioral interviewing techniques, which seek to illustrate a candidates competency and compatibility for a new job by matching the patterns of performance and styles of work from their past to the ideal for the position. This theory, in human terms, says that as you grow older and gain experience, you simply become a more refined version of the person you always were (a key point in Marcus Buckingham's new book entitled Go Put Your Strengths to Work).
If you are interesting becoming a better forecaster and trend observer, you'll like this post showing Paul Saffo's (Institute for the Future) Rules for Forecasting.
Enough philosophy. How do you forecast? How do you take out the bias, but ground your forecasts in reality? Are complex methods better than simple methods? How important is an understanding of the business (thus the passion that would lead to bias and to errors) or is it better to have a "pattern recognition" person that doesn't know the business? Your insights and experience greatly appreciated.
Posts
