Week 10 discussion!

Hey there!

Last week discussion got pretty active, so I thought it be nice to open the floor one last time before the final.


My week 9 results were pretty ugly, as I made model selections based off my own wrong charts.
Fixed it, and this week, satisfying results, 8th place.

Sharing feedback:


I guess the heat map is what one would be aiming for, heavier concentration in the top left.
Didn’t get any claims above 10k$. Can’t complain.

Is there any way to know the actual industry distribution of claims in those 4 groups?
None, 0-5, 5-10, 10+

Moreover, something encouraging, is that off the 20 top insurers, have one of the biggest market share (along with @guillaume_bs) which hints that I have room to increase prices. Or is this only luck? :thinking: :shamrock:


One thing’s certain… I used up all the 10 weeks to train models and didn’t use that valuable time to tune the pricing. In hindsight, that was a mistake. No more time to tweak the pricing :compass:

Will have to do a hail mary for the final. :rocket: :moneybag:


And you? :wave:

  • How are you feeling going into the finals?
  • Confident?
  • Conservative, or going for a hail mary?
  • Would you like another mid-week leaderboard, recycling previous weeks data?
    **wink wink at the organizers :laughing:


Good luck all! :trophy:


My week 9 had 0 profit loading

and I introduced a profit loading for week 10

I changed my approach a lot and wasted many weeks, but I feel like I could turn a profit on the final leaderboard. Will it be big enough? I don’t know.


I’ve been trying to figure out where I am on the classic price optimisation curve:

Last week I adopted a ‘Gucci’ style pricing strategy, applying a huge profit loading and ending up with a profit for the first time, as well as a pathetically tiny market share.

This week I reduced prices, which helped improve market share a little and overall profit a lot. At 1.5% market share I’m probably still too expensive, but if I reduce prices again I could end up overshooting the optimum, if I haven’t already passed it.

The probability of winning an individual policy keeps changing as people update their prices, and the profitability per policy keeps changing depending on where the large claims happen to fall each week, so the optimal pricing point keeps shifting. It feels like there are too many moving parts to figure out an optimal strategy, so a Hail Mary might work as well as any other strategy. I’ll probably just change prices a bit, allow for features only relevant to the final challenge, and hope for the best…


Same model, reduced profit loading -> higher market share but slightly negative profit

Hmm… I think I will tweak my profit loading a bit again