This post is going to be a sort of introduction to some risk analysis that I’m going to try to incorporate into player analysis and forecasting. I think we have a long way to go when it comes to the risk management side of player evaluation. I’ll admit that I don’t have all the answers when it comes to risk analysis, but I’m going to try to incorporate some qualitative and quantitative features so we can better understand what kind of risk we are dealing with. Baseball HQ uses a reliability feature to serve its purposes of risk management. I’ve written elsewhere that I’m not a big fan of this tool. Baseball HQ has admitted that it’s struggled in various fantasy leagues throughout the year using its risk strategy.
Thatrisk management plan sounds good in theory, but I believe the reason it didn’t work out was because it wasn’t projecting future risk as much as describing past risk. I’m going to try to take the first approach and project future risk. While you may not agree with the conclusions, I hope you find the process acceptable. If so, you can plug your own figures into the process and come up with your own risk assessment of a player. These are some of the tools I’m going to use to try to do this:
When we try to measure risk, we are really concerned with two areas: probability and impact. Probability is the chance of an event happening; impact is the magnitude of the event if the risk were to occur. These will be measured using the following scales:
Very low: unlikely to occur
Low: may occur occasionally
Medium: as likely as not to occur
High: likely to occur
Very high: almost certain to occur
Very low: negligible impact
Low: minor impact
Medium: notable impact
High: substantial impact
Very high: threatens success of player
We can then quantify these scales and combine them to create a sort of risk table:
The numbers in the squares are found by multiplying the probability numbers by the impact numbers. You can adjust the scales toward your liking, so if you worry more about impact you can have a multiplicative scale instead of a linear one like the one above. We can then use these colors to give player’s low, medium and high risks and give an overall risk level for a player. The qualitative risk areas I am going to analyze are based partly off of Baseball HQ’s past risk assessments of players. These areas will be:
Experience: How much past data do we have on a player? More data can give a more certain projection. Also, I do not weight minor league experience as heavily as major league experience. Some past research has shown that minor league equivalencies are not as accurate in projecting major league performance as major league data.
Playing time: How certain are we that a player will gain consistent playing time? This does not factor in playing time that could be lost to injury.
Age: Players can drop off suddenly as they age, while some experience large improvements in performance when they’re younger.
Burnout: Will a player have a sudden, catastrophic injury?
Skill risk: This last category is inspired by chaos theory. A part of chaos theory is the butterfly effect, which means small changes in one area can produce large, non-linear consequences somewhere else. For players, this means a small improvement or decline in one or two skills can produce a large improvement or decline in his overall performance.
Two quantitative tools I will use are stats generated by PECOTA and a reliability score given in Marcel projections. PECOTA produces breakout, collapse and beta statistics. A breakout score is, put simply, the chance a player will have a large improvement in offensive performance while a collapse score is the chance a player will have a large decline. Beta measures the volatility in a player’s projection, with one representing average risk, below one being below average risk, and above one being above average risk.
The second tool I will use is the reliability score given with Marcel projections. It shows how much regression is used when coming up with a player forecast. If a player has reliability score of 0.8, his performance was regressed 20 percent to the mean. I’m going to try to show how we can use these tools to forecast risk by using Troy Tulowitzki as an example.
Troy Tulowitzki was taken seventh overall in the 2005 draft by the Colorado Rockies. His minor league numbers weren’t mind-blowing, but he needed only 517 minor league at-bats before reaching the major leagues for good. Tulowitzki had an excellent first full season in the majors and finished second in Rookie of the Year voting to Ryan Braun. Many felt Tulowitzki actually had the better season once the huge defensive differences between the two were factored in.
Tulowitzki’s 2008 season has not been as impressive, though. He went through a miserable slump to start the year before tearing his left quadriceps tendon on April 29. Tulowitzki would not return to the big league team until June 20. On July 5, Tulowitzki suffered a cut on his right hand after he slammed his bat in frustration from being taken out of a game. He went on the DL again, but returned on July 21. Since then, Tulowitzki has been playing much better, bringing his season lines to .250/.324/.388. While this is nowhere near his stats from his rookie year, he has had to deal with two injuries. He could have varying positions on draft lists next year. Let’s look at how risky a player Tulowitzki is for next year.
Qualitative risk assessment
Overall Risk: Medium
We have a pretty good sample size for Tulowtizki’s performance. He should have around 1,200 major league plate appearances by the time he is done with the 2008 season. However, I put his risk impact at high because we’re not sure how much of the real Tulowitzki we have seen. His performance this year could be a mirage because of the injury or it could be a warning sign that Tulowitzki may have peaked early.
Probability: Very low
Impact: Very low
Overall risk: Low
Tulowitzki is signed to a long-term contract, and the Rockies clearly want him to be their shortstop for the present and future. Even when he was struggling early in the year, the Rockies stuck with him. He doesn’t have much of a threat behind him to take playing time. Overall, Tulowitzki should be a safe bet to see almost all the at-bats at shortstop for the Rockies barring an injury.
Probability: Very low
Overall risk: Low
Tulowitzki will play the 2009 season as a 24-year-old. Players this age tend to not have a severe collapse. People may make comparisons to another Long Beach State shortstop: Bobby Crosby. While their injury problems have been quite similar, Crosby really didn’t hit a wall until he was 26. Tulowitzki may hit that wall later, but I think he is a low-risk proposition in 2009 for an age-related decline. It’s also worth noting that there may be some upside left in Tulowitzki. He’s at the age where we should expect some improvement, and there’s a small chance that he could see a large improvement in skills.
Overall risk: High
Part of the risk with Tulowitzki comes with his playing style. Tulowitzki plays all out all the time. Because of this, he’s a rather high risk of suffering an injury. We’ll have to wait and see if he’s one of those guys who goes on the DL at least once a year or whether he can learn to control himself a little better.
Overall Risk: Medium
Tulowitzki has made great gains with his plate discipline this year. He’s walking a bit more and striking out a lot less. His hit rate has been down this year, but I think we can mostly attribute that to bad luck. The only real concern I would have with him skill wise would be with his power. He doesn’t hit fly balls at a tremendous rate and hasn’t shown a great home run rate over his two years. Overall, his batting average skills look good, but the power could be a question mark.
Overall risk level
Medium, but on the lower side of that. We could call it a low yellow I guess. I find this by taking the numerical overall risk, found by using the chart above, for each category, adding them up, and dividing that by five. I then see where that number falls on the risk chart.
Quantitative risk assessment
Breakout: 29 percent
Collapse: 20 percent
PECOTA saw Tulowitzki with having about average risk in his forecast. However, it gave him a relatively high chance of taking his game toward the next level but also a decent chance of taking a step back. I’d say that next year Tulowitzki would probably have a solid breakout score and a lower collapse score. He should get a pretty high reliability score next year from Marcels.
Overall risk assessment
Overall Tulowitzki is a medium-level risk, moderate upside player. Most of this risk comes from his playing style and the injury possibility that results from that. His power also could be considered a moderate risk. I would say, though, that Tulowitzki is not as big a risk as some might think after his tough year. If you are one of those willing to embrace risk, he might be worth taking a shot at in your fantasy league next year. However, if you are more risk-averse, you do not have to discount Tulowitzki’s value too much.
This is the first in what I hope is a series of risk profiles. I’m hoping we can take risk management to the next level when it comes to player forecasting. I’d love to hear your thoughts and suggestions on this topic.
A good primer on risk management, and where I found the risk table, can be found here.