Subject: Chat with Andrew Stein
From: Andrew Stein <steinlink@gmail.com>
To: barriticus@gmail.com

me: HEY DUDE LOOK WHAT YOU'RE MISSING IN TEXAS
me: http://www.okcupid.com/profile/HighPocketsinTX
Andrew: http://cdn.crooksandliars.com/files/uploads/2008/09/tom_delay_mug_shot-790518.jpg
Andrew: nearly indistinguishable
me: jesus christ
me: well done
Andrew: you're in dallas?
me: yep indeed
Andrew: got your own apartment?
me: yep
Andrew: swell
Andrew: did I ask if you used chrome?
me: yeah, and you were going to show me something but I had to leave the coffee shop, don't have internet at my place yet
Andrew: brooklyn.ego.fm:8000
Andrew: it takes about 10 minutes to boot though
Andrew: on the server I mean
Andrew: because it is very complicated
me: link's not working for me
Andrew: yeah, i know
me: going to "oops"
me: ok
Andrew: i jus tstarted the server
Andrew: it takes abotu 10 minutes to boot
Andrew: so it will be up in 10 minutes
me: oh, gotcha
Andrew: but I have to admit, its among the best of my work
Andrew: website that predicts stock prices by AI analyzing twitter chatter
me: how long have you been working on this?
Andrew: few days
Andrew: weekend project
me: have you tested it?
Andrew: just on friday
me: what's the next step?
Andrew: picked BP and YHOO heavily, both rose
me: seems like something you might want to keep secret
Andrew: not necessarily
me: remember PI?
me: you'll accidentally find the name of God
Andrew: good analogy, but im already afraid of jews
me: at any rate, what's the game plan with this?
Andrew: well, i wrote it to get a job
Andrew: got fired on thursday
me: got fired?
me: why?
Andrew: politics
Andrew: no big deal, i have cash saved and got a nice severence
Andrew: but i need a new job eventually
Andrew: heres the thing
Andrew: its mainly a toy - taking investment advice from twitter is a bad idea in general, too volatile
me: right
Andrew: though it does seem to be really good at picking stocks that are "in the news"
me: one sec
Andrew: thats not super hard to do by hand
Andrew: k, its up now
Andrew: its filled with fake data right now because the market is closed - so im not entirely sure how china got in there
me: how does it differentiate between "Yahoo HQ blown up" and "Yahoo invents perpetual motion machine," or does it?
me: is chatter deemed to be on average an indicator of near-future gains?
Andrew: regression analysis - it associated words and phrases with a score by analyzing past tweets and assessing their correlation with price flux at certain intervals
Andrew: yes, by the hour
Andrew: the picks are "hour picks"
Andrew: going to cut it to 20 minute picks when i get more twitter feeds
Andrew: shorts are "sells", longs are "buys"
me: right
me: so how long are you set for financially up there?
Andrew: depends on how much i spend
Andrew: december?
Andrew: but i want another job
me: I imagine you can find another job before then?
Andrew: i have a few offers right now
Andrew: so likely
me: very good
Andrew: i was thinking that this project might be able to turn me a profit on subscription
Andrew: I think i can get reddit headlines for a few hours - then charge stupid daytraders $100 a month for more granular data from teh system
Andrew: then it doesnt matter how accurate the data is
Andrew: i can market it as "automated social network data analysis for data traders"
Andrew: ie, if you want to invest in yahoo for a few hours as a trader
Andrew: you can click a button and see in depth detail on what the internet is saying about yahoo right now
me: $100/month is low
Andrew: possibly - theres a price point that is reasonable in there somewhere
Andrew: and its a great turnkey business
Andrew: all i have to do is collect to payments
Andrew: also program occasionally, but i do that for free
Andrew: im going to run on tuesday and prolly all next week to test data
Andrew: then potentially launch "ad supported" version on monday
Andrew: if it makes waves, ill do a subscription part
me: well done
me: give me the word and I can promote it
me: when you're ready
Andrew: i just want to make sure it runs well on at least one full day of data
Andrew: it still does weird thing sometimes
Andrew: i had to remove ticker symbols like "a", "all", "how", "see", "via", etc
Andrew: because they make lots of false hits
Andrew: eventually, the algorithm should be able to distinguish which of those are tickers and which are english noise