back to rpi forecast

The Forecasts are updated daily.

Here are the steps in detail:

- Update all of the wins and losses to date
- Using Jeff Sagarin's "PREDICTOR", calculate the probabilities of winning for every remaining game
- Draw random Wins and Losses based on these probabilities for every remaining game
- Figure out the end of season RPI for every team based on the completed and simulated wins and losses
- Sort the RPIs (numbers between 0 and 1) to get RPI ranks (counting numbers, 1, 2, 3, etc.)
- Save the details from this one simulation
- Repeat the simulation 10,000 times
- Calculate the Expected RPI and Expected RPI Rank, etc., by using my 10,000 simulations

The end-of-season RPI forecasts on this page include all past and future opponents in a team's Strength of Schedule and as such should give a better measure of what the RPI will look like on Selection Sunday than the day-to-day RPI. After all, that is what really matters.

Including future opponents only makes sense if you realize that the future opponents may be better or worse than all past opponents and adjust your winning percentage accordingly. I accomplish this by calculating probabilities of winning for each future game by using data from past games. Once I have calculated these probabilities, I determine the most likely outcome of all future games. This is the most important part of the RPI forecast.

How do I know this? I have backtested my model for previous seasons and found that the Root Mean Squared Error AND the Mean Absolute Difference in predicting the Rank is lower for the RPI Forecast than for the RPI itself. Also, now that I have a season (2006-07) under the belt, we can look at those results too. In forecasting the RPI, I included only games that were on the schedule at the time, so that means that up until the last week or so, no Conference Tournament games were on the schedule. So, with that in mind, here is a graph showing the ability of the RPI Forecast at predicting the end-of-regular-season RPI versus the ability of the daily RPI to do so. What you see below is the Mean Absolute Deviation of the RPI Forecast and today's RPI and the MEAN Absolute Deviation of the daily RPI and today's RPI plotted over time as the season progressed. The Mean Absolute Deviation is the the average difference between the RPI Forecast or daily RPI and today's RPI. It just tells you how far away you expect the average forecast to be. Lower is better.

As you can see, the RPI Forecast did better than the daily RPI at predicting the end-of-regular-season RPI. The diference is much better early on in the season, when the average error for the daily rpi is about 80 and the RPI forecast is about 40 spots away on average.

**Expected RPI Rank:**This is what I expect the RPI rank to be at the end of the season. It is probably what you came to my webpage for.**Overall:**This is the one most likely end of season rank outcome. Because the Expected RPI Ranks are not usually round numbers, I added this variable which is basically the rank of the RPI Forecast (which are the numbers between 0 and 1. See below) and not necessarily of the Expected RPI Rank. The Expected RPI Rank is still a better measure of an individual team's future RPI rank, but if you want to see standard ranks (1, 2, 3, etc.) then you may like this more.**Team:**The name of the basketball team's school. Click on this to look at a graph of the Expected RPI Rank forecasts over time, and a histogram of today's simulations.**Conf:**The conference of the basketball team's school. Click on this to look at the time series of RPI forecasts by conference.**RPI Forecast:**This is what I expect the RPI (not the RPI rank, mind you!) to be at the end of the season. The RPI Rank is more important.**SOS Forecast (rank):**This is a forecast of the Strength of Schedule for each team. The SOS is basically your (2/3) * Opponents' winning percentage + (1/3) * Opponents' Opponents' winning percentage. The rank is just a sorting of the SOS number.**Current W-L:**This is the CURRENT record of the team versus Division 1 opponents.**Projected W-L:**This is what I expect the team's D1 record to be once they have played all of their games on the schedule as of today. It is based on probabilities of winning which are derived using all games to date.**1-25 W-L:**This is what I expect the team's D1 record to be against the top-25 RPI teams at the end of the season.**26-50 W-L:**This is what I expect the team's D1 record to be against the 26-50 RPI ranked teams at the end of the season.**51-100 W-L:**This is what I expect the team's D1 record to be against the 51-100 RPI ranked teams at the end of the season.**101-200 W-L:**This is what I expect the team's D1 record to be against the 200-336 RPI ranked teams at the end of the season.**200+ W-L:**This is what I expect the team's D1 record to be against the 101-200 RPI ranked teams at the end of the season.**OOC W-L forecast:**Just like the Projected W-L, except restricted to Out of conference (or non-conference) opponents.**OOC RPI forecast:**This is the RANK of the forecasted OOC RPI. The OOC RPI is just the RPI applied to non-conference opponents.**OOC SOS forecast:**This is the RANK of the forecasted OOC SOS. The OOC SOS is just the SOS applied to non-conference opponents.**Current RPI:**This is the RANK of the current day-to-day RPI. This is the traditional RPI, calculated using only games-to-date.**Overrated:**This equal to: Expected RPI Rank - Current RPI. It is a measure of how overrated I think the team if you rely on the traditional day-to-day RPI. Teams with negative overrated values are underrated.**t-stat:**This is equal to: Overrated / (Standard Deviation of RPI Rank Forecast). This measures how many standard deviations away the current RPI is from the Expected RPI Rank.

**Rank:**This is the rank of the projected RPI for the conference.**Conference:**The name of the D1 basketball conference. Click on this to look at a graph of the RPI Rank forecasts over time.**RPI Forecast:**This is what I expect the RPI (between 0 and 1) to be at the end of the season.**SOS Forecast (rank):**This is a forecast of the Strength of Schedule for each team. The SOS is basically your (2/3) * Opponents' winning percentage + (1/3) * Opponents' Opponents' winning percentage. The rank is just a sorting of the SOS number.**OOC W-L forecast:**This is what I expect the conference's Non-conference D1 record to be once they have played all of their games on the schedule as of today. It is based on probabilities of winning which are derived using all games to date.**pctg.:**This is the OOC W-L forecast expressed as a winning percentage.**OOC RPI forecast (rank):**This is what I expect the RPI (between 0 and 1) to be at the end of the season. The rank is the sorting of the OOC RPI forecast.**OOC SOS forecast (rank):**This is the Out of conference Strength of Schedule forecast for the conference. The rank is the sorting of the OOC RPI forecast.

**Expected RPI Rank:**This is what I expect the RPI rank to be at the end of the season. It is probably what you came to my webpage for. It is the average rank from all of the simulations.**Team:**The name of the basketball team's school. Click on this to look at a graph of the Expected RPI Rank forecasts over time, and a histogram of today's simulations.**Conf:**The conference of the basketball team's school. Click on this to look at the time series of RPI forecasts by conference.**Standard Deviation:**This is The Standard deviation of the RPI Rank forecasts. It is calculated from the simulations.**Min:**This is the minimum (or closest to 1) RPI rank from the simulations for the team. You can think of it as the best case scenario for the team.**2.5th %ile:**This is the 2.5th percentile RPI rank taken from the simulations. It forms the lower part of the 95% confidence interval.**10th %ile:**This is the 10th percentile RPI rank taken from the simulations.**25th %ile:**This is the 25th percentile RPI rank taken from the simulations.**median:**This is the median or 50th percentile RPI rank taken from the simulations. If the distribution is skewed, which it is for very good and very bad teams, it might be a better measure of expected RPI rank. Half of the simulated ranks were at least as high and half were at least as low as this rank.**75th %ile:**This is the 75th percentile RPI rank taken from the simulations.**90th %ile:**This is the 90th percentile RPI rank taken from the simulations.**2.5th %ile:**This is the 97.5th percentile RPI rank taken from the simulations. It forms the upper part of the 95% confidence interval.**Min:**This is the maximum (or closest to 336) RPI rank from the simulations for the team. You can think of it as the worst case scenario for the team.

back to rpi forecast