Introducing Expected Contract Value – Part 2

Introducing Expected Contract Value

Part 2:  Inputs & Outputs

Part 1:  Justification, Theory, & “Contract Analytics”

As we described yesterday, Expected Contract Value is an objective metric that enables valuation and comparison of contracts, as well as team salary cap budgeting, by using regression analysis to identify the influence on team-decision making of the relationships among various contract characteristics. Today, we will describe both the inputs and outputs of the metric.

Save:Cap: This input is a ratio of (i) the amount of cap savings that the team would realize upon releasing a player to (ii) the player’s cap number if the team does not release him. A team may be more enticed to save $1 million in cap room by releasing a player who will count $3 million against the cap than it would be by releasing a player who will count $10 million against the cap. Furthermore, a team may be dissuaded from releasing an overpaid and underperforming player if doing so would result in a larger cap hit than refraining from doing so. This input can be thought of in the following way: How beneficial (or detrimental) would it be from a salary cap perspective to release this player?

Cap:Avg: This input is a ratio of (i) the player’s cap number if the team does not release him to (ii) the average per year (“APY”) of the player’s contract. Even though a central tenet of Expected Contract Value is that the face value – and by extension, the APY – of a contract is irrelevant, it seems fairly apparent that it does matter with respect to how teams and agents assign values to players when negotiating contracts.[1] Because of this reality, a team may be more inclined to release a player whose cap number has climbed above his APY, as he may now be deemed “overpaid” (in terms or cap dollars). On the other hand, a supposedly underperforming player may have wiggle room before being released if his current cap number is less than his APY. A player’s cap number doesn’t mean much without context; APY provides context to how talented or valuable a player purportedly is. This input can be thought of in the following way: Is this player accounting for a greater portion of our salary cap than is justified, given the value that we have previously assigned to him?

 [1]To the extent that Expected Contract Value succeeds in changing this dynamic, the regression results, and the corresponding Expected Contract Value formula, will change over time.

Save:Avg: This input is a ratio of (i) the amount of cap savings that the team would realize upon releasing the player to (ii) the APY of the player’s contract. This input approaches issues raised in the previous inputs from a different perspective, accounting for certain scenarios that the others don’t. For example, a player’s Save:Cap might be very high (meaning the cap savings are very beneficial), but if the player’s cap number in that season is less than his APY, then the Save:Avg will be lower. A team might not care if it would save a large percentage of the cap space taken up by the player if they feel that they are allocating a “bargain” amount of cap space on the player. This input can be thought of in the following way: Would the salary cap benefits (or detriments) of releasing this player be outweighed by the degree to which the player is currently a bargain (or albatross) in comparison to the value that we have previously assigned to him?

Contract:Complete: This input is a ratio of (i) the number of years of the contract that have been completed to (ii) the total number of years of the contract. This input accounts for the “optionality value” of NFL contracts.[2] Because NFL contracts are generally not guaranteed, NFL teams have a “team option” on most contract seasons that is indistinguishable from team options that NBA or MLB teams have on some contracts (with dead money comparable to “buy-outs” in the contracts of the other leagues). According to finance theory, the option itself has value. A team may be less inclined to release a player in year four of a six-year contract than in year six, because it needs to at least entertain the possibility that the player will improve or rebound his performance in year four, thereby preserving the optionality value in years five and six. In year six, there are no future options to worry about, so this consideration would disappear. This input can be thought of in the following way: How much of an opportunity exists for this contract to generate surplus value in years beyond the current one?

 [2] For more on this topic, see Optionality in NFL Player Contracts, Eugene Shen, AllianceBernstein (eugene.shen@gmail.com)

Age:Peak: This input is a ratio of (i) the number of seasons the player has played to (ii) the number of seasons that a player in the theoretical peak football player age has played.[3] The input is expressed in terms of seasons rather than actual ages so as to place the ratio on the same order of magnitude as the other inputs. A team may be more inclined to release a player considered to be in his decline phase, as the prospects for a drop off in performance would be higher. This input can be thought of in the following way: Does the player’s age make it more or less likely that he will fail to justify the amount of salary cap space we would allocate toward him for this season?

 [3] We realize that “peak age” has never been demonstrated with any degree of conclusiveness in the context of the NFL, and that it could theoretically vary by position. Overall, using this input slightly increases the accuracy of the regression model, but the input does not have a large effect on the outputs until the age becomes extreme (34+).  This is primarily an issue for kickers/punters who play into their late 30’s and early 40’s.

Delay:Dead: This input is a ratio of (i) the amount of potential dead money upon a release that could be deferred to the following year if the release is done as a post-June-1st release to (ii) the total potential dead money upon a release. While a large amount of potential dead money (affecting Save:Cap and Save:Avg) may deter a team from releasing a player, the team may still be willing to release the player if a large percentage of that potential dead money can be deferred to the following year. This input can be thought of in the following way: To what degree can the potential detrimental salary cap effects of releasing a player who otherwise “deserves” to be released be deferred until the following year?

At this point, one might question why we have not included the players’ position as an input. The theory of Expected Contract Value does not dictate that the position of the player would matter. We are of the opinion that team decision makers would view player value – however they determine value for the purposes of evaluating players – as equal between positions. X Units of Value is equally valuable coming from one position as compared to another position, assuming that the Units of Value are calculated in reference to some sort of replacement-level type of concept that takes positional scarcity into account (such as how WAR does in the context of baseball; 1 WAR from a catcher is the same as 1 WAR from 1 first baseman, but it is easier for a catcher to accumulate 1 WAR due to positional scarcity). And if NFL teams are not utilizing such a measure of value, then we fail to see how they would even go about altering their decision making process based upon which position the player whose contract they are evaluating happens to play. Remember, the idea of Expected Contract Value is to forecast what decisions a team will make regarding a contract, not to determine how much “value” the contract provides to the team (as WAR does).

So, either (i) teams possess the ability to measure value,[4] in which case such measure of value would have to account for position in order to be useful[5] and position should therefore not be “double counted” in forecasting decision making via Expected Contract Value, or (ii) teams do not possess the ability to measure value, in which case they would have no non-arbitrary method or reason by which to base decision making on the player’s position. The obvious exception to this latter option is if the position-related decision is made on the basis of the specific team’s needs, which is exactly the sort of unpredictable variable that Expected Contract Value intentionally avoids utilizing (remember, the accuracy is 80% even while ignoring considerations such as this).  Having said all of that, we acknowledge that decisions regarding starting quarterbacks present unique considerations that probably render Expected Contract value much less applicable as to those contracts.

[4] For these purposes it does not matter whether the measure of value is accurate; it only matters that the teams utilizing it believe that it is accurate, and therefore base their decisions at least in part on the outputs of the measurement.
[5] An on-field value metric that did not account for position would not be useful. If a team had X cap dollars to use, and it utilized different metrics for different positions with no way to translate the metrics into a common denominator, it would have no way to prioritize its use of the X cap dollars.

As previously mentioned, these inputs are run through a regression analysis to determine the degree to which each of them has correlated with the binary output of whether or not a player has been released in given contract years for historical contracts. The result of this regression analysis is a formula into which the same inputs can be entered for new/current contracts. The output is a likelihood, between 0% and 100%, of the player being released in each year of the contract. The output looks as follows, using 2014-offseason Eagles signee Riley Cooper as an example.

Riley Cooper
YearExpected Outcome
201499.9%
201596.4%
201661.3%
201739.8%
201826.2%

The percentages in the right-hand column represent the likelihood that Cooper will remain on the Eagles for each year of the contract. While many Eagles fans have called for his release due to an underwhelming 2014 season, Expected Contract Value finds that to be very unlikely.  He is still likely to remain under contract in 2016, unlikely to remain under contract in 2017, and very unlikely to remain under contract in 2018.  Remember, the Expected Contract Value formula has no idea whether or not Riley Cooper is actually good at football.

It is important to keep in mind that Expected Contract Value provides a forecast, not a prediction. Think of it like a weather forecast. The meteorologist on the television in the morning might say that there is a 70% chance of rain on that day. But at the end of the day, the clouds in the sky will either have produced rain, or they will not have produced rain. If the meteorologist were making a prediction, he would need to pick either “rain” or “not rain.” But what he is doing is providing an expectation based on a comparison of that day’s weather patterns as compared to the weather patterns of previous days. And despite having an incredible amount of data to generate a great sample size, as well as highly sophisticated tools, the meteorologist is still frequently “wrong” with his forecast. Such is the nature of forecasting. Expected Contract Value has shown to be “correct” in its forecasting approximately 80% of the time (in terms of rounding the expected values up/down to 100% or 0% in order to make a prediction). We think that meteorologists would have a much better reputation if they could match that rate of success.

Another way to think of Expected Contract Value is like a baseball performance projection system, such as Nate Silver’s PECOTA. Generally speaking, PECOTA compares the attributes of current players to the attributes of similar past players, and then projects future on-field performance for the current players based on how the similar players progressed throughout their careers. PECOTA utilizes many more variables, and it projects many more outcomes, but ultimately Expected Contract Value is projecting the “performance” of the contract (in terms of a “yes” or “no” outcome each contract season) in a similar way by comparing the characteristics of current contracts to the characteristics of past contracts and then seeing how the past contracts “performed.”

Back to the Riley Cooper example. In the next chart, we will add in Cooper’s salaries in each year of the contract in order to generate his Expected Contract Value.

Riley Cooper
YearSalaryExpected OutcomeExpected ValueAdjustment
201499.9%$5,000,000
2015$1,000,00096.4%$964,000$3,000,000
2016$4,500,00061.3%$2,758,500
2017$4,200,00039.8%$1,671,600
2018$4,750,00026.2%$1,244,500
Subtotal$6,638,600$8,000,000
Total$14,638,600

In this case, an adjustment is needed due to the fact that the entire base salary (as well as a signing bonus) was guaranteed in 2014, and $3 million of the 2015 base salary is guaranteed. Expected Contract Value still thought that there was a 0.1% chance that Cooper would be released in 2014,[6] but even if he were, he would have still banked the money. Likewise, only $1 million of Cooper’s $4 million base salary for 2015 is jeopardized by the 3.6% chance that the Eagles release him.

[6] The likelihood will never be 100%, which makes sense when you consider crazy things can happen along the lines of Ray Rice and Aaron Hernandez, or even team decisions such as with DeAngelo Hall in Oakland.

As you can see, the Expected Contract Value of this deal for Cooper is just $14,638,600. Compare this to the $22,500,000 face value of the deal. The “guaranteed money” of this deal is $8 million, which is clearly under-inclusive of value. The “three-year payout” is $13,500,000, which is also under-inclusive of value.

The APY of Cooper’s contract is $4.5 million. We explained yesterday that a problem with APY is that it assumes all contract money has an equal chance of being earned. Having used Expected Contract Value to generate expected outcome likelihoods, we can now utilize an appropriate type of APY. In this case, it would be $2,927,720 (the Expected Contract Value, divided by the number of seasons). We can call this “Expected APY.” Generally, we think that Expected Contract Value is a more useful metric than Expected APY. It is a counting metric that takes into consideration relative likelihoods, and in general “more counting metric” is better than “less counting metric,” regardless of the rate at which the counting metric accumulates. However, for players who anticipate signing another significant contract at the conclusion of their current contract, the Expected Contract Value may be less important than preserving a good Expected APY while at the same time ending the current contract as soon as possible so as to sign the following contract.

One last important thing to note about both the inputs and outputs of Expected Contract Value is that each contract season is analyzed as its own snapshot in time, as of the beginning of the relevant offseason. So in the case of an option bonus that is not fully guaranteed, the contract season in which the option bonus would be paid is analyzed as a comparison from the perspective of (i) the option bonus being paid, if the player is not released, versus (ii) the option bonus not being paid, if the player is released. In subsequent contract seasons, Expected Contract value assumes that the option bonus was paid in the previous year, and the effects of that are accounted for. Likewise, in the case of extensions, only the new money and new contract seasons are considered for the purpose of APY related inputs, but the effects of the previous contract (prorated amounts, etc.) are taken into account in all other respects.

In Part 3, we will engage in contract comparison to demonstrate how Expected Contract Value highlights the effects on value of structuring seemingly similarly sized contracts in different ways. In Part 4, we will demonstrate how Expected Contract Value can be used to aid teams in budgeting their salary cap structures. In Part 5, we will address feedback received throughout the week.

Expected Contract Value was created by Bryce Johnston and Nicholas Barton.

Bryce Johnston earned his Juris Doctor, magna cum laude, from Georgetown University Law Center in May 2014, and currently works as a corporate associate in the New York City office of an AmLaw 50 law firm.  Before becoming a contributor to overthecap.com, Bryce operated eaglescap.com for 10 NFL offseasons, appearing multiple times on 610 WIP Sports Radio in Philadelphia as an NFL salary cap expert. Bryce can be contacted via e-mail at bryce.l.johnston@gmail.com or via Twitter @eaglessalarycap.

Nicholas Barton is a sophomore at Georgetown University. He intends on double majoring in Operations and Information Management and Finance as well as pursuing a minor in Economics. Currently one of the leaders of the Georgetown Sports Analysis, Business, and Research Group, Nick consults for Dynamic Sports Solutions, an innovative sports start-up that uses mathematical and computational methods to evaluate players. He also writes for the Hoya, Georgetown’s school newspaper, and his own blog, Life of a Football Fan. Nick can be contacted via e-mail at njb50@georgetown.edu.