ModelOff 2019 - Round 2: Turning it around

Description
This is the only ModelOff round that decides who makes it to the finals and who doesn’t. I had a very good feeling coming to this round – I was well-rested and well-prepared. After the first look at the questions I saw that they suit my strengths to a great extent. I always tend to do well with data cleaning questions (Case 3 – It’s a Mess). You can’t imagine how messy real-life client data sometimes could be.

As for Case 4 (Angry Utterances) I did quite well with a similar word search case in New York finals 2017. On the awards ceremony that year, there were 2 finalists the organizers praised for their approach to solving this type of a word quiz. Amazingly, one of them was myself while the other one was Diarmuid Early, now member of Question Design Team, who actually was the author of the Case 4 question. Anyway, this year I didn’t get a change to fully enjoy Dim’s talents at designing questions and here is why.

Section 1 (Rapid-Fire models) and Section 2 (Turning it Around) went smoothly for me. BTW, Case 2 was very similar to Hard Times from R2 2013. Spent an adequate amount of time on both of these and had 1:34 left before the start of Section 3.

Section 3 turned out to be a mess for me however. This case dealt with cleaning data of the number of zoo visitors per animal in 2018 and 2019. The data was purposefully presented in a complete disarray. The data for 2019 was cleanable and I spend not too much time on it. However, cleaning data for 2018 turned out to be a nightmare.

Getting the dates and animal names was comparatively easy. However, getting out the number of visitors for each animal was tough. I realized that there are various approaches to search for the number of animals, however, each of them was OK only for a part of the data array.

I decided to do data cleaning step-by-step. E.g. use one approach and be able to clean ~1500 rows. Then use another approach on the remaining rows and clean some other 1000 rows. Then repeat until all the data is cleaned. That was my mistake, as it turned out that there are more than 10 different approaches, and each of them was quite time-consuming.

Eventually I solved the case, but I spent most of my time on it. Right after the competition my mind offered me another approach that would probably be much faster. Will try to solve the case with this approach when I have more spare time. Maybe I could even publish the solution here if ModelOff organizers agree.

So, at the end of the 3 hour timeframe, I had about 5 minutes left when I started case 4. Decided just to open it and at least try to guess the correct answers. But it turned out that Case 4 did not provide questions – you had to type in the correct answers! OMG, that was so frustrating! However, I didn’t give up and still tried them. Amazingly, it turned out that the first 2 questions under case 4 (that gave as much as 5 points, BTW) were very easy. Managed to solve both of them within the 5 minutes remaining.

So, to sum everything up, I didn’t get it to the finals this year, but I was very close. Will definitely keep participating and will probably do some more preparation for ModelOff 2020. I also can’t wait when the promised monthly competitions start – these should be a great fun for the financial modeling community.