I’m not complaining, given the current state of affairs
My understanding of the rules is as follow:
I’m competing for N policies in a market of 10 competitors
then for another N policies in a different market of 10.
for a total of kN policies in K markets. (are k and N known?) (do we get to know our market share sin the k markets?)
To qualify under the participation rule, I have to get a market share of 5% in at least one of the k markets. Is this correct?
One of my worries is that if one persons predicts their claim using the “baseline mean model” they will probably get a 50% market share, selling to everyone with an above-average risk. This leaves *an average market share if 50/9 = 5.5% for the other competitors, so it’s pretty hard to qualify everyone in that market.
If I understand this correctly, this might be an issue because “baseline mean scripts” currently represent more than 10% of the scripts on the leaderboard. An average market will get 1 of these scripts.
Should “baseline mean scripts” be ignored from the profit competition? I’d love to adverse-select them to bankruptcy, but not if it means I don’t qualify for the participation rate rule.
Also, there is probably also a person in the market with with “predicted_premiums = predicted_claims” and that means there is not much market share left for people who are trying to earn a profit.
I could also be way off, I’m just scared the 5% rule may go awry.
huge thanks again for all your work.
I think the “5% rule” is a bit different : you need to get one policy in 5% of the markets :
" Participation rule. Your model must participate (i.e. win at least 1 policy) in 5% or more of the markets it is placed in."
So the 2 examples you gave do not create any issues with this rule.
Still I would also be very curious to know more about our market performance (market size, market share, offered / written premium & margin, …) to adapt our margin strategy !
You’re right, my reading comprehension wasn’t up to par.
I guess the question that remains is “how many policies per market” because 1 per market is not the same in a 10-policy market and a 1000-policies market.
Knowing our market size/share is definitely needed, here’s to hoping we get access to it.
Firstly, I have just posted an announcement that may be of interest.
Secondly I will address you questions below:
How many policies in each market. Every market has exactly the same N policies that you compete for. The only thing that changes from one market to another, are the other 9 competitors that you are playing against.
The 5% participation rule. As @guillaume_bs states, this rule requires that you get at least 1 policy in 5% or more of the many many markets that you are placed in. In otherwords, you must have nonzero marketshare in at least 5% of markets.
Baseline scripts. I hope (2) has alleviated those worries, but if not, rest assured that the mean model baseline does not satisfy the
non-zero non-negative profit rule, so they would already be disqualified from the markets without ever entering the equation. However, some may have tweaked this to make it work, in which case, well, that is their model.
Predict_premium = predict_expected_claim, what then? Yes some people might do this, but with market randomisation their effect will be mitigated. With time, and especially after this weeks’ leaderboard, more and more players will also realise that this may not always be profitable market strategy for everyone (market conditions notwithstanding).
Feedback about market performance. Yes, this will likely be present from next week, see my comments here.
And… congrats on the first leaderboard!
thanks again @alfarzan. Let me summarize again, just to be sure:
Let’s say there are a total of 10 000 quotes to be priced this week.
a) Will they be split evenly in K markets of N quotes so I bid on each quote once (K * N = 10 000),
b) Or will there be K markets, all consisting of all 10 000 quotes so I bid on each quote K times(K * N = 10 000K)
I guess my question is “If I am underpriced for that policy with the 100,000$ claim am I going to sell it multiple times?” My understanding from your reply is “b”, but I learned not to trust my reading comprehension
Yes it is (b) indeed. You price the same contracts multiple times. So if you are underpricing a very expensive claim than you are doing so many times. But each time there will be other players that you are competing against.