Can math help you win IPL?
Recently, I accompanied a friend to the sprawling campus of IIT Madras with an unlikely subject in my mind: Sports. You don’t usually connect an IIT with sports. It might boast of play grounds and gymnasiums, and you might even have spotted its students jogging on its roads (passing by signs that ask motorists to go slow to give way for deers) or dribbling around in its basket ball courts.
But, you always get a sense that sporting ability is not a source of pleasure at IIT, academic excellence is. Scour the index of Sandeepan Deb’s book IITians : The Story of a remarkable institution and how its alumni are reshaping the world, and you won’t find a single reference to sports – there’s no football, no basketball and no cricket. IITians it would seem are generally more skilled between their ears than between cricket stumps.
I said sports was in my mind, but really, it was the maths part of the game. My friend and I were there to meet Rahul Marathe, an assistant professor at its department of management studies. Rahul, along with two BTech students Bharat Bansal and Tarun Inani, has developed a model to find out the right price for IPL players based on their performance in the first four editions. This in effect means using maths to compress their cricketing ability into one number – the dollar amount you should be bidding for in an IPL auction. Ahead of the fifth edition of IPL, they compared the numbers their model produced with actual amounts different teams paid for their players and found that in some cases the teams paid too much and in some cases, paid too less.
"You can use statistics", Michael Lewis explained in an NPR interview soon after his book Moneyball was published, "to sort of dig below the surface of baseball and find the hidden game, find attributes, for example, in players that are very important but not highly valued in the marketplace, and also find attributes in players that teams pay a lot for that actually aren't worth that much when it comes to victory and defeat."
That pretty much defines what Rahul and his team are trying to do – except that this model doesn’t go to the level of detailing (that Moneyball does) and it certainly hasn’t found its Billy Bean. “It’s an academic exercise right now”, says Rahul.
One way to test if the model is good is to ask if it could have predicted – if not accurately, at least with some amount of certainty – the results of IPL that ended two days ago. Aggregating the value of the players, the model placed Chennai, Mumbai, Kolkata and Delhi at the top. And they indeed were. It showed Morkel among the three most undervalued players, and he ended up taking most wickets. (Yet another undervalued player Bravo, helped CSK win at Eden Gardens against KKR). Rahul found that players from abroad are usually undervalued.
The model promises to get better as it gets more data and as it uses more attributes. It’s somewhat limited at present, says Rahul. ‘A player’s age is probably an important factor in IPL, and we haven’t considered that for our analysis’. Listening to Rahul, we felt he had more questions than answers right now – but underlying his approach was a sort of confidence that you can use math to improve the outcomes for your team.
That probably comes from his background. Rahul has been teaching operations research for MBA students and for the undergrads since 2007. He got his PhD from Iowa State University, and was doing his post doctoral research and handling a couple of classes there when he saw a recruitment ad from IIT Madras. He had never been to Chennai before, but Iowa State had enough students from the state to make him feel a kind of affinity of Chennai. (“Ask me about any Tamil movie that hit the screen between 2000 and 2006!”) He loves running and IIT-M’s huge, green campus came as a big bonus.
What he loved most, however, was the conversations he had with a colleague G Srinivasan, who now heads the department of management studies at IIT. Like Rahul, Srinivasan found great pleasure in applying maths in different fields, for example, using ‘data envelopment analysis’ to find out if Rahul Dravid was a more efficient player than Sachin Tendulkar. (As it turns out, Dravid is).
When the first IPL auction in 2008 took everyone’s breath away Rahul was not among the first to train his OR guns on the auction numbers. Others did. But using data from one-day matches – leave alone test – on IPL format did not appeal to Rahul. This format demanded a different set of skills. By 2011, however, he started working with data from four editions to basically see if he can assign a numerical value to the players based on their batting and bowling records – along with Bharat and Tarun – and eventually assigned a dollar value to those numbers based on the money teams were actually spending for players. When Bharat presented the paper at a seminar at IIM-Kolkata last December, it won him a prize.
Still, the model needs some more tweaking before it can be tried in IPL auctions. The model does not take into account the value that comes out of the permutations and combinations of players in a team , the synergy that comes out of having three or four players. It also doesn’t give any additional value to the captian, who could turn out to be a key differentiator in several matches. Even more importantly, a team owner might want a player not so much for his cricketing ability as for his charisma and popularity, which might be important for a team like say Royal Challengers or even Chennai Super Kings – since the brand can extend to other products as well.
Rahul and his team are working on these – aware of the complexities involved in such an exercise. I asked Rahul if he had seen the movie Moneyball. No, he said. “I told myself I won’t see the movie till I finish the book. I haven’t finished the book yet.”