Alright folks, here we go again! This year I'll be releasing valuations for a few different formats (# teams, # players per team) so you may find the values more relevant to you. Youâll find the player values relevant to you in this mammoth cheat sheet, or by using this spreadsheet, or this dashboard.
Iâve employed the same methodology as previously described, although espn.com haven't released projections for GP (games played), FGA (field goal attempts), and FGA (free throw attempts) this year so these forecasts are based solely on those provided at FantasyPros this year where available. In cases where FantasyPros didnât project GP for players, Iâm using the average of GP across the league to forecast how many games that player will play. These player valuations are more relevant to auction drafts, but you can still find my valuations for use in snake drafts. Below are the final player valuations for 10 team / 10 players (per team) leagues - which assumes that each of the 10 teams in the league are looking to draft 10 players.
It's very tight up the top this year with very little separating Curry, Harden, Davis, and Durant. This decision comes down to where the value lies in auction drafts (whether you can get any of these players for less than they're worth) or personal preference in snake drafts. Also, players with a zero value are basically on a level playing field and can be used to complement various combinations of players put together using funds from the budget. The recommendation for these âfillerâ players would be to not spend any budget on them, although in reality you will have to fork out at least 1 unit of budget for each of them.
Finally perennial sleeper pick Paul Millsap seems to be getting the attention he warrants, which means he may no longer come cheap. Jeff Teague and Al Horford look like the value buys in Atlanta. Mike Conley looks like he could be a bargain over in Memphis and one to also keep tabs on, while Gordon Hayward and Kris Middleton are also worth jumping on if others in your draft sleep on them.
On the flip side there are some players that look to be highly overrated at this point, and the recipe for that seems to be dominant big men who can't make free throws. Although Gobert, Whiteside, and Drummond certainly have their merits, they seem to be getting far too much credit (as far as fantasy value goes) and their blocks and boards come at the detriment of other categories. Kevin Love is another notably overrated player heading into the season.
The pairs plot above shows the relationship between all of the projected statistical categories, as well as the final player valuations (in the 8 team / 10 players format). Of particular interest is the final row, which shows fantasy value to be intrinsically linked to all categories except rebounds (REB) and blocks (BLK). This is likely due to the damage these players cause to other statistical categories from a fantasy perspective based on the current pool of talent in the league, and goes some way to explaining why the methodology finds some of the aforementioned players to be substantially overrated. It might be worth considering a punt in both of these categories this season. Metrics related to scoring points have the strongest relationship to value this season in the standard 8-cat format.
Somewhat surprisingly (to me at least) the projections for Rondo have come out very dull this year, rendering him effectively useless. If you feel that he'll play better than what has been projected of him he could be worthy of a gamble.
Finally, what's a new season without rookies? There are some enticing new additions coming into the league this year that look capable of making an immediate impact on the court (and more importantly the stat sheet amiright?). Look for those who will see plenty of court time in their first year in the league.
Itâs almost time for the new NBA season which means NBA fantasy drafts are just around the corner.
After finishing 9th (out of 14) teams and missing out on the playoffs in my league last year I decided that a few refinements were needed. Last year was a good learning experience, and I feel that a lot of those learnings can be implemented in the valuation process to refine the player draft values. This year Iâve made a few tweaks to the way the valuations are calculated, in the methodology as well as incorporating another source for player projections. Once again Iâll be trying to uncover the best value pickups in this yearâs draft based on comparing my valuations to ESPNâs - and for consistency the valuations in this post will be based on 10-team leagues with 13-man rosters, playing the standard 8 categories (in-line with ESPNâs valuations).
As I havenât deviated substantially from the methodology I outlined in last yearâs valuation piece, Iâll only cover the key updates that were made this time round, before focusing on the results.
The key changes since last year are:
- Moved the process to across to R, to facilitate more iterations and allow for valuation variables (such as roster size, and league size) to be parameterised.
- Incorporated a secondary source (FantasyPros) for player projections, and blending those projections with ESPNâs.
- Last year I aimed to draft 13 value picks, however I soon realised how difficult trades were to execute and the value of the waiver wire. So this year I aim to draft 10 value picks, keeping 3 roster spots open for pickups from the waivers.
- Correction of final player values based on projected number of games played to account for players starting the season injured as well as injury-prone players.
And now itâs time for the results. To provide a reminder of what the valuation process looks like over all of the iterations, here is a recap of the process for a few selected players.Â
After these iterations are completed a correction was applied to the player values to produce the final valuations as represented my cheat sheet (at the end of this article). The following chart shows the final draft values of the top 100 players, in groups of 10, from the top 10 through to 91-100. Now that the player valuation process is complete, we can analyse the differences in these player values in comparison to ESPNâs, in order to identify this yearâs value picks.
Kevin Durant looks like a steal if people in your league are silly enough to let him slip for cheap. In fact, according to my analysis ESPN has undervalued many of the top players, while Paul Millsap (a longtime underrated favourite of mine) and Michael Carter-Williams look like potential good value picks. Â
One final piece of advice - make sure you have a clear idea of who youâre going after prior to the draft so you can keep your eyes on them and set aside your funds to go after them. If you want a cheat sheet to reference on draft day, you can find one here.
I would wish you luck, but you shouldnât need that anymore!
March Madness has to be one of the most interesting opportunities each year for prediction, I mean even Obama gets involved. This year Iâve designed a model to try to predict every possible match-up in this yearâs tournament as part of the Kaggle March Madness competition. Each of the 68 teams could theoretically play each other team, which means that there are 2278 different potential match-ups that could take place. However it's important to note that these games would be played at various stages of the tournament. Clearly a teamâs performance in the tournament has a large bearing on how likely they are to win their next game, so Iâve had to factor this into the model. This is one of the major reasons as to why filling in a perfect bracket is worthy of a billion dollar prize - it is far more difficult to pick the result of every match before the tournament starts than it is to pick every result prior to tip-off.
In order to incorporate this important aspect into the pre-tourney predictions Iâve had to project the events that would have had to have taken place in order to land a team in the particular match-up being forecasted. So if weâre forecasting the potential Albany NY vs Arizona match-up, which could only take place in the championship game we need to incorporate the information required for this game to take place. Namely that Albany NY not only won its wild card game but also its games in the round of 64, round of 32, sweet sixteen, elite eight, and final four, which would certainly require taking care of some heavy-hitters. Which is why even though I have projected Albany to have a slim 3.5% chance of upsetting #1 seed Florida in the round of 64, if they can their probability of winning their next game increases considerably to 27.7% (although this will vary based on who they end up playing) with the knowledge that theyâd have to have beaten Florida to land themselves in this game.
The below table shows the pre-tourney probability of a team winning itâs match-up each round:
The percentages listed in the table state the probability of a certain team winning their match-up in that particular round, given that theyâve reached that stage of the tournament. For instance, I have projected that Virginia have a 97.1% chance of winning their first round match.
Arriving at these probabilities requires considering each possible match-up for each team in the given round and the projected likelihood of the team winning that match-up. These likelihoods must also be weighted by how likely those respective opponents are to make it to that stage - as outlined by the projected progress table:
According to my model, Arizona are the favorites to win this yearâs tournament although they only have a 14.5% chance of doing so. Next in line are Florida at 10.2%, Wichita St. and Virginia each at 7.1%, and Creighton at 6.7% round out the top 5. In terms of which conference the winner will come from I have the West on top with a 30% chance of taking home the title, followed by the Midwest with a 24% chance and the East and South both with a 23% chance. It's looking like this March is going to be as hard to pick as ever.
Itâs that wonderful time of the year again, where NBA fantasy seasons are kicking off to put an end to the productivity of everyone taking part. This year marks my first serious fantasy league - the league has 14 participants who will all be taking it seriously because each person has $50 on the line. Itâs also going to be my first experience with an auction draft, which means I need to do some research. So effectively my productivity has already taken a hit even before the season has tipped off. Iâm sure Iâm not the only person, as the draft is probably the single most important event of the fantasy season. Finally, the last of my firsts is that this year marks my first foray into the head-to-head fantasy format. So far Iâve spent some time preparing myself for draft day, so I will cover that work here rather than the day-to-day team management here.
One of the most important realisations Iâve made in my experience with fantasy basketball is that the strength of your team is highly dependent on how the other teams in your league are set up. Itâs important to have a team that takes advantage of key weak categories in your league, which also means that the value of certain players can be highly influenced by what key areas of weakness exist within your league. Due to this, Iâve set up a trade engine that reflects each players value by considering the current teamâs projected performance (against the other teams in the league ) before and after the proposed trade in question. This valuation engine will continue to be helpful throughout the season as a means of evaluation trades. Some trades can be beneficial to both parties, and these are the trades that are important to identify. Unless you want to be a troll and waste everyoneâs time (including your own).
I soon realised that I could also build a valuation engine for the draft to be used to draft the best team for a once-and-done approach, but once you consider that you can trade players throughout the season itâs clear that success in the draft should be measured by the (fantasy) dollar value of the team youâve drafted. Even if your team isnât properly set up immediately after the draft, having valuable trade pieces enables you to attain a better team in the longer run via a few smart trades. So the question changed slightly from âwhat is the best team I can set up given how the other teams in my league are set upâ to âwhat is the most valuable team I can afford?â. Winning the draft now becomes more about investing wisely, and the best drafter is the one with the most bargains. It is also going to be key to use up the budget completely, so you donât necessarily want to wait too long to get a team full of steals. You may want to score some stars, and fund them by surrounding them with bargain sidekicks - but all of this can be summarized in two main rules:
1. Donât overpay for anyone on your team.
2. Spend as much of your budget as possible.
If you manage to achieve both of these, you will have had yourself a very successful draft.Â
The most important piece of information you need to achieve this is a fair valuation of each player and Iâve tried to tackle this problem by means of simulation. Using the standard (because Iâm cheap like that) ESPN cheat sheet valuations as my starting point, Iâve simulated 1000 affordable combinations of players and then compared their projected performance. Iâve then isolated the top 5 percentile (top 50) of teams based on their projected performance relative to one another and compared the ratio of teams including each player in the pool of 1000 to the proportion of teams featuring that specific player out of the top performing teams. For instance, if LeBron (valued at $70 initially) was a member of 50 of the 1000 of the randomly generated (affordable) teams - he might have then gone on to feature in 5 of the top 50 performing teams. LeBronâs feature ratio for the randomly selected pool would be 5% (50/1000) and his feature ratio in the top performing teams would be 10% (5/50). Such a result would suggest LeBron was under-priced at that particular valuation, and his valuation would increase (along with all the other undervalued players) by $1 at the end of that iteration. The over-priced players would receive a reduction of $1 in their valuation (provided they werenât already valued at $1) while the fairly priced players wouldnât see a change in their valuation. This details one iteration, and the process was repeated 50 times. The rational for only needing 50 iterations is that Iâm assuming ESPN are not off by more than $50 in their valuation of any player. After the 50 iterations are complete we can analyze the valuations at the end of each round to ensure that a level of stability was reach after 50 rounds, and if so we can be fairly comfortable that the players are fairly valued. This method is neat because it accommodates for a few things that are very difficult to quantify - such as the âreplaceabilityâ of certain players who excel in rare categories as well as the number of teams in the league to some extent.
Hereâs an illustration of the player valuations over the entire 50 rounds:
It is clear that some playersâ values have changed significantly from their starting point (or relative to ESPNâs valuations) while others have stayed relatively stable over the process. Notice that James and Durant stand atop this list and are relatively inseparable in terms of fantasy value, but more importantly their values seem to have plateaued well before the final valuation. This is true for most of the players that have been valued here with the exception of some stellar players who are starting the season injured (Bryant, Westbrook and Rondo in particular). Because Iâve only considered per-game stats, productive players that wonât feature for a large part of the season will be severely overrated as the projections assume all players play the same amount of games. I didnât address this issue because my league is set up in the head-to-head format, but it would be a key (and simple) consideration to make for rotisserie leagues. That said, these player valuations should be reflective of value in rotisserie leagues too.
Other things to consider are that Iâve only looked at the top 200 players (according to ESPN projections). I felt 200 players was about right as my league will consist of 14 teams or 182 players. The 18 extra players provide a little flexibility for teams to be constructed in different ways based on player preferences within the league. Clearly injured players need to be considered more carefully, mostly for how much of the season theyâll be contributing for. I gambled on Rose last year, and it didnât pay off - so if youâre drafting an injured player be aware of the inherent risk. I havenât factored these considerations into my valuations so beware that any injured players will be substantially overrated here if they donât play near a full season of basketball. The final caveat to these valuations is that Iâve assumed the league is set up with the standard categories (FG%, FT%, PTS, REB, AST, BLK, STL, 3PM).
Here are the 50 most valuable fantasy players (based on their final valuations):
Again, Bryant and Westbrook are overvalued here due to their injury statuses. The other thing the chart above illustrates is that their are some players that ESPN have overvalued (most notably Love, Gasol and Ibaka) and many that have been undervalued. For the most part the players here have been undervalued, but by varying amounts and so the true player value can be thought of in terms of how much each player has been undervalued. That most of the players here are shown as undervalued is due to the number of teams in my league. Whereas ESPN's draft valuations are based on a league of 10 teams, mine are based on a league of 14 teams which creates more competition for the better players.
Also see the valuations of players valued 51-100,101-150 and 151-200.
Another thing to keep in mind is that it might not be easy as it sounds to spend as much of your budget as possible, especially if youâre in the business of making wise trades. Some players are likely to be (at least close to) accurately valued by other participants in your league, and others will be wildly under-or-overvalued. If you bide your time for too long waiting to pick up a team full of thoroughly undervalued players, you will have a valuable team but chances are you will also have a lot of money left over. That left over money is a waste. What you want to ultimately do is draft the most valuable team based on the playersâ inherent values. To do this it is helpful to have an idea of how much other people are likely to be willing to play for certain players. Now that weâve already fairly valued all the players we can look at what combinations of players perform best, and this can provide an indication of what strategy to pursue in the draft. We can easily compare the new valuations to the starting point and see which players were gravely undervalued by ESPN (and perhaps many other players in fantasy leagues if theyâre going by those ratings) and identify which players could/should be prime targets in this yearâs draft. The easiest way of illustrating this is by looking at the differences between my player valuations and the ESPN valuations. The other consideration is to look at the combinations of players that provide you with the most valuable team under the assumption that ESPNâs valuations provide an indication of what price youâre able to secure them for. Although you probably wonât be able to get all of these players for those same prices, it still offers a useful indication of some team options you might want to consider. For instance, this team was the most valuable affordable team (values based on my valuations and affordability based on ESPNâs).
This alone demonstrates how my opinions have changed on just about every single one of the players on the list with the exception of Durant (we all knew he was stellar).Â
Finally Iâve put together a cheat sheet with some information to help you out in this yearâs draft. Follow these tips and youâre bound to end up with a solid fantasy team, and with some luck you might be able to really stick it to your friends this year.Â
In-game win likelihoods based on time remaining and point differential
It's not at all mind blowing that the time remaining in the game and the point differential at that point are key predictors of the outcome of a game. But it can still be interesting to look at different game situations and see how those games unfold to get an idea of what an unassailable lead really is. I've often written off a team down 15 points in the 3rd quarter only to later find out that I've missed a thrilling comeback victory. Those are the games I do not want to miss, so let's see if we can do something about that. This piece of analysis should also be able to provide a benchmark as to just how remarkable different comebacks are and allow for comparison.
To get a feel for how most games develop we can take a look at the point differential of all games in the season at different stages of the game. What we'll look at next is the distribution of the point differential at different stages of the game to get a feel for how most games develop. The density plot below illustrates the distribution of in-game scenarios over the season:
The first chart represents the density of games observed with a particular margin at various points in the game. As expected the majority of games are relatively close, with slight skewness in favour of the home team suggesting some home court advantage at play. This might be more clear focusing on the bottom chart which represents a cross-section of the first chart at the end of regulation. The point differential at the end of regulation follows a fairly normal distribution, skewed slightly towards the home team's favour (negative away point differential). The colour scale shows the Z-score of the frequency, with the most frequent points represented by a warmer colour. It's also interesting to note the frequency spike at A_diff = 0 - which are games which have ended tied. This makes sense due to a team's incentive to try to force overtime in end-game situations - from what I've noticed a team down two points will generally go for a two point shot, whilst a team down three near the end has no option but to throw up a three-point attempt.
 The next step is to look at the probability of win for these same combinations of point differential and remaining regulation time based on the results of similar games. The chart below shows the likelihood of an away victory via the colouration of the points in the scatter-plot, while the points themselves represent a specific game scenario. Intuitively the likelihood of victory approaches a definite level as the score margin blows out and the end of the game draws near.
Here's how these simple win-likelihoods look over the course of a game when considering only time remaining and point differential as factors:
Interestingly the above diagram shows the Nuggets had essentially locked in the win by the end of the first half - the win likelihoods are represented by the thin dotted lines. At this stage, the model is still too raw and more input variables should be considered to give more robustness to the projections. Finally we close by looking at these likelihood models in a closer game where there are multiple lead changes.
Betting markets seem to be the benchmark of predictive performance for an analysis of this type, so one possible test measure could be to retrospectively look at whether this model could have been used to take advantage of mispricing in in-game betting markets. So that's more data sourcing to add to the list...
Now that the data is set up, it's easy to extend this analysis to look at a couple of other interesting areas - call it a bonus. The first of such is looking at 4th quarter comebacks over the 2009/10 season, by isolating all games where a team trailed with less than 12 minutes remaining in the game and looking at the proportion of these games they managed to win.
It looks as though OKC was by far the least resilient team over the course of the season when it came to mounting a home comeback in the fourth quarter. In fact the biggest deficit they were able to overturn was 3 points, this was perhaps a sign of just how young the team was at the time. On the other hand Orlando look to be a team that could never be written off even when trailing late at home, while Cleveland, Atlanta and Denver won most of their home games in which they trailed by either 1 or 2 points in the final period.
Interestingly OKC were clearly a more resilient team at away from home, but still their inability to mount a comeback from more than 5 points down in the final period suggests that overall they still weren't a very resilient bunch. The standout poor performers (in terms of comeback-ability) were the Bucks, Timberwolves and Nets. The most interesting team here are the Kings, who although they weren't really able to win a large proportion of games when they were close (primarily because they weren't a very good team), their likelihood of winning didn't drop when they faced larger deficits. As far as character goes, this team might have had one of the more never-say-die attitudes. Who knows, that might happen when you're used to finding yourself down big.
Finally, let's look at the most impressive comeback(s) of the season - the (s) because this was actually a joint-prize with two equally unlikely comebacks. One was an away win (New Orleans) after trailing by 7 with 3:55 to play and the other was another away win (Phoenix) after trailing by 5 with 1:30 to play.Â
The Swish Cheese is a home for basketball discussion and analysis, although I realize most people won't actually find my site after being redirected by Google to something related to swiss cheese, but I only want the people who endeavor to get here anyway.Â
On the back of finishing second in my fantasy basketball league this season Iâm ready to share some of my expert knowledge. On a more serious note, Iâve got a couple pieces of analysis that I want to get up on here in the future.
Iâd like to hear your ideas too, and not just limited to basketball because at the end of the day Iâm interested in analytics first and foremost. So be sure to give me some feedback, and drop me a note if you ever have an interesting idea that youâd like to explore with me.
the swish cheese @theswishcheese - Tumblr Blog | Tumgag
