Back to Community
Advice for starting on an algorithmic options trading project

Hello everyone. I am a MS in CS student at UMass Amherst. I am working on a project to automating algorithmic options trading. I was hoping someone could advice me on how I should start the project, what literature i should look at and any algorithms I could check so as to understand the process better. I would be really grateful. Thank you so much.

7 responses

Are there any guidelines on this project? Are you restricted to an algorithmic strategy with just options? If you have the choice, it would be easier to create a algorithmic strategy based on traditional long/short equity trades. I helped a friend who is working towards a PhD in CS that had a similar project. He created an algorithm to backtest the dividend discount model (DDM) and the relative strength index (RSI).

No I do not have any guidelines as such. The plan is to design a very simple benchmark that would almost be mechanical and then try to create an algorithm based pipeline. Yes it has to options. It should be possible to apply equity based knowledge to options, right? Or at least understand the process.

Gotcha. Getting hold of historical option data may be a challenge. I haven't kept up with Quantopian for about a year, but I don't believe they have added options to the list of their services. If you go to a large university, there is a good chance you have access to Wharton Research Data Services which may have historical option data (I can't quite remember everything that's included in a WRDS subscription).

Once you get the option data, you could create a Black-Scholes pricing model to signal whether the options are priced appropriately. Be careful though, this can quickly become a very complicated. It may be easiest to run the algorithm on just the SPY or a few ETFs as opposed to the entire universe of US equities.

That sounds good thank you so much Graham. What do you think would be a good benchmark I could design to test my algorithm against?

Option strategies typically involve leverage so it depends on the context of your algorithm. For the sake of this project, benchmarking it against the S&P 500 total return is probably okay.

  1. If you're a grad student looking to create first model in options, then having a positive PnL is good enough (and tough) target for more than a year trading time. Unless you planned to keep buying calls on AAPL over last 10 years or you planned buying puts on VXX in 2017.
  2. In options one typically makes money in following ways:
    a) sell puts when ImpliedVol is high. This is to profit from premium. Buy calls if ImpliedVol is really really low and you expect the underlying to go up.
    These are normally manual trades. Note the involved risk/exposure.
    b) continuously trade the volatility just for a cent or two. I left this for last because it is tough and this is where most of the automated strategies work. For this, you can generally assume your buy/sell price = (bid + ask)/2 +/- some_slippage.

  3. note the well known stats (and I might be off a bit on this) : 90% long options (i.e. option buyer) loose money by expiration. This doesn't mean 90% of first-buyer of an option looses money (LOL).

good luck,

Check the tastytrade channel and quantconnect. The first does a lot of "research" on options strategy and the second provides a framework to test options strategies.