Hacker News new | ask | show | jobs
by arthurcolle 2032 days ago
I use a hybrid complex event processing system that is mostly custom built, as well as a large scale data ingestion framework that takes in around 2-3GB of price data and other relevant information every day. It took a really long time to get this part of my setup working because without a decent way to kick off a process that subscribes to this data easily, everything else is going to be a huge pain in the ass.

I put on trades on a 2week-1 month time frame, and aggressively sell winners and use stop losses to get rid of shitty positions. I use GARCH, ARIMA models, and have recently been playing around with some more exotic ML techniques, but I've found that momentum trades work the most reliably.

I'm working on a SaaS API to try and make some money off of my data collection systems, but that's still in the works.

EDIT: I try to identify interesting individual options that are, according to my model "mispriced" given a few signals. I seek out multibagger options. Most recently, I purchased the $PLTR 11/27 $22 strike calls two days ago, and made a killing (was around a 6x gain). If you are methodical in your data collection processing, there's a lot of valuable information that can be obtained. It just requires lots of data munging (which, honestly, is so painful all things considered). But turning 1k into 10k is a wonderful experience, especially if you can reproduce it.

I have taken the well known 30-40 advanced options strategies and created a dynamic programming algo to basically run through every single listed security and every reasonable combination of options that would yield these strategies, with some filtering to make it computationally tractable, and then computed what the collateral costs would be (for spread trades, for ex.). That creates a reasonable doable list, and then I run a straightforward backtest. From here, I either pursue a trade or not. I like the idea of fully automating this stuff but its just not there yet. In any case, systematizing a lot of the boring crap is a great way to more deeply understand the landscape. You're never going to compete with execution-based firms but you can definitely take a view based on underlying fundamentals/macro landscape and then tune a slightly longer based system appropriately.

7 comments

At the risk of a naive question, may I ask where you get your option feeds from?

I've been hesitant to pull the trigger on any of the "professional quality" feeds given that I've been treating this sort of thing largely as a hobby, but the less pricy options seem highly variable in data quality, granularity and availability. I guess if you're trading on week/month timescale this may not be necessary, thus my musing.

Thanks for sharing, sorry if this is probing more deeply than you'd be willing to broach.

I source option data from a combination of some paid APIs, some free sources. You can find a lot of data sitting around the internet if you're careful about it and respect robots.txt... Some people even pay for XIgnite and other crazy expensive data sources, but then have not secured their own endpoints (some SaaS operators have done this) so I scoop that data up too.

but as @marketgod suggests, you can definitely just use broker APIs, thats going to be the best bet for the most part and if you're price sensitive.

Get a ThinkOrSwim account with $10 in it and you get their real time API. You can create a socket and store the entire chain. Perfect data does not exist in this world, anyone who claims they have it, has a lot of money.
Taking a stab in the dark here but let me know if I am close. You are using GARCH and ARIMA on the options pricing data to identify moves in the market. Once you identify a move you try out a bunch of different combinations of buying/selling puts/calls. You then take a look at the possible positions and see which wouldn't require an absurd collateral and execute on it?

For something like PLNT how did you run a backtest? It seems like there is only ~1 month worth of data.

How does one learn about finance enough to do something like this?
I worked at Goldman for a few years. I would suggest doing something like that.
Can you elaborate what roles would be preferable when working at place like Goldman? What typical qualifications and experience are expected?
They shouldn't be much different from other engineering jobs. One thing they will look for is "business acumen". Learn all the finance buzz words, pay attention to the market, and credentialize. Finance people seem to love prestige and credentials (just my experiences). A degree which includes finance is good. You can also get a CFA foundations certs. A CFA charter would get you real respect, but I think they have experience requirements for that.
How does one learn enough about finance to do something like that?
This sounds super cool. I was eyeing PLTR since it went public. I missed my chance and didn't pull the trigger. Glad it worked out for you!
Why do you think you missed it?
Do you collect historical OHLC data?
Yup, use a few different APIs to paint a picture going back around 20 yrs, including fundamentals, filings, ex-div dates, splits, reverse splits. Currently focused on US equities exclusively, and the foreign companies that have secondary listings on US exchanges. Definitely keeping me busy, just with those ~7668 tickers.
For what period, daily or per-minute candles? I have OHLC market close data going back to 2000.
would you be interested in taking outside money to run on your algo?
Hi, thanks for asking - I'm definitely open to that in the near to mid-term. I'm currently working to scale the systems and also expose my robust APIs online as a SaaS, but I'm interested to hear what you're thinking.

Feel free to reach out to me at the email listed on my profile!

wow, Love this!