Chloe gave me a CD of nine Iron & Wine songs on Sunday before we went to my hockey game, and I haven't taken it out of the player in my car yet. I don't know all the songs, and I actually don't even listen to all of them, but some of them are extremely beautiful. Each Coming Night is one such song, and the lyrics above are from Passing Afternoon, which almost made me cry the first time I really listened to the lyrics and let my thoughts go with them.

Today at work I discovered that I can listen to This American Life despite the censors. I had always tried to get there by going to thislife.org, which is their official address which they advertise, but today I tried just seeing if going through a Google back door would work. And of course it did. It also turns out that you can just type in "thisamericanlife.org" to get there. Similar to how they've blocked "Gmail.com" but not "mail.google.com." Anyway, I listened to the episode about Georgia rambler, except that the first half-hour was something I'd heard a long time ago, and for some reason the last ten minutes didn't download. I also listened to the one called "Origin Story," and I was hoping for a bit on elevator pitches, but that never seemed to come up...

In another attempt to find meaningless high-paying work, I looked up pay rates for toll booth operators. Possibly enough to support me living alone but not enough to justify doing it. Reverting to "own a laundromat" as the default.

I don't have much left to do at work, and so I don't do too much at work, and I kind of feel guilty but also bored out of my mind. I finished way more than my manager expected me to this summer and so now all I have left is a report that I have to give next Thursday, and he's on vacation pretty much for the next four days and all I had to do was edit some reports... which I already did. I decided to give Radio Free Asia another try, but it's Khmer broadcast is nowhere near the sound quality of Voice of America. Which is unfortunate, because RFA is a way more respectable institution than VOA, and they also have thirty more minutes of broadcasting every day. Another downside is that they don't stream their audio, you have to download it.

Two kids at work, Dave and Han, work together on the ninth floor. They always talk about how they've done almost nothing all summer, and it makes me a little jealous. Dave just reads the NYT webpage, and Han sleeps or studies for his actuarial exam. Apparently their boss is way too busy to give them any work, and nothing at all had come across their desk in the last four weeks. And they still got paid! How cool is that! They just got a day or two worth of work because their boss realized if he didn't do it soon, he'd never give them anything... so he passed them off to someone else to give them work.

As for what I did this summer at work... Basically, started out with a whole crapload of data that was separated into policies that our company had sold as insurance, and huge databases of information on companies and ZIP codes. Both files had a common field, so we could link them up by policy number, and so then we had a ton of information about every company we gave a policy to. Well, almost every company. We found out some basic information about the business so far (loss ratio, average GWP, project types, etc.) just to understand what we were working with, and then put all the data on 2009 terms so that we could analyze it fairly. Then we run "banding" on every variable... so basically, we just take every policy and put it into one of ten buckets depending on where that variable falls on the continuum. Each of the ten buckets should have roughly 10% of all the policies. So now we've got like 800 variables or so and we have all the policies into the buckets, so we put them into modelling software that lets us see in a really nice visual format what's happening to the loss ratio as the variable changes. The meat of the project is the next step, going through the 800 variables and seeing which ones are predictive and consistent, and which ones are not. Basically, if the modelling software puts out a diagonal line, you're in good shape. If it puts out anything else, you can move on. There were lots of diagonal lines, but not all of them are consistent across time and state and company size, etc. So in the end, we narrowed it down to about eight variables, and then we tried a bunch of models with different combinations of those variables included or excluded until we found one that appeared the most predictive and consistent, so that included six of the variables. From there, we go back through the data and project what would happen to our profit if we had used the loss ratio information we got based on these new variables to price our insurance better, and then try to convince our underwriters to act on this new information.

Does that make sense? We find lurking variables that affect insurance scores and then use them to price insurance better.

I played hockey again tonight and found out that this kid on my team named Kozi is going to PSU main this fall. He's really good, so I told him he should play on a team with me. I'm trying to put together a bigger team, actually, and he'd make a really good addition.... we do lack a goalie, though. Hm. I'll figure that out.

Peace out.

posted by Singing Farewell 11:12 PM on Tuesday, August 3, 2010