Friday, March 14, 2014

03/14/2014 sector update

The sector model sold XLP and bought XLU before the close.


  1. XLU marched (40.70 to 40.89) while XLP stagnated (42.56 to 42.58). Would the roundtrip swap be considered a losing flipflop?

  2. That was actually a false signal. The end of day data called to stay in XLU. We picked up an extra 1% in January from a switch, and lost about .5% this week. It pretty much evens out.

  3. Interesting, so the real time data 15 min before close reversed at the close? Can you tell how close to the close the real time signal reversed back to XLU?

  4. Also, the part I don't understand - if the 3/13 close showed XLU still in the lead and XLP being a false signal, why didn't you trade back to XLU on 3/14 open but rather do it at the 3/14 close?

  5. It was .33% of a difference by the end on my model.

    There was actually a .14% favorable gap first thing in the morning -- but I have an infinitely better investment called a good job :).

    As I said, I gained an extra percent in January and lost half a percent in March. Sometimes the slippage works in my favor and sometimes it doesn't. Since it tends to even out I don't sweat it.

  6. To add a little more confusion into the mix, the person managing the Sector Fund initiates the trades at 1:45pm based on the calculation the model has at that time.

    The time is in a 11am to 2pm window in which the trades are free for the clients.

    His own reporting of returns is actually slightly higher than the end of day model that I run in parallel. We did this so that we could make sure his trading window didn't cost clients on returns, and so far it's been slightly favorable.

    We're both up over 10% year to date, and pleased with the results so far.

  7. I'd be mostly interested to know whether using EOD trading (just prior to close) or at 1:45pm is statistically significantly better than trading at the next day's open (based on previous day close) to warrant adherence to real-time state of the model, including factoring of extra trade friction due to whipsaws.

  8. The opening gaps sometimes go in your favor and sometimes don't.

    In any case, the Sector Fund being managed by Gold Coast Advisors LLC only trades at 1:45pm. Those trades are free, so there is no friction.

    My personal account running in parallel trades at 3:45pm.

    The charts that are automatically generated by the model show end of day data.

    We've seen no statistical difference between any of them.

    I also have a fourth version of the model that avoids some of the whipsaws, and there is no statistical different there either.

    The PROBLEM someone will have isn't the occasional whipsaw, but the core trades themselves if they are having to pay for trades and their account is below a certain size.

    Let's say there are 10 trades a year (a good average for the model), and let's say the model outperforms the S&P by 15% each year (it does better, but let's stick with that). In a typical etrade account you'll pay 10 dollars EACH way on the trade, so that's 20 dollars for the round trip on those 10 core trades.

    Now we are at 200 dollars a year in trading costs for an etrade account. If your account were only 1300 dollars to begin with you'd lose your entire outperformance in trading costs.

    Hence the Sector Fund. The trading window of 11-2 allows for free round trip trades. Since there ARE no trading costs it doesn't matter whether you trade 10 times a year or 100 times a year. If you have 800 dollars in the account or 800,000 you'd get the exact same return rate.

    In any case, we'll continue to run these in parallel, and I'll report the Sector Fund's returns compared to end of day returns. But so far the only statistical difference is the advantage the Sector Fund has with free trades.

  9. I should have clarified that by friction, I mostly mean market impact of the trade itself. XLU, for example, traded 2.7M shares today at ~$41. At 1% threshold of volume for market impact of an intraday trade, that's 27K x $41 = $1.1M which is, well, a very small fund. So there, liquidity in my humble opinion is rather limited for commercial deployment of this system.

  10. Oh -- we've thought of that. Would be a good problem to have! The scalability problem can be resolved in a number of ways:

    1) We have an alternate version of the model that has different gauge settings and trades on different days with basically the same returns.
    2) We can add more sector based etfs from the subcategories.
    3) We can trade in multiple daily segments with real time (instead of near real time) data.

    And we can use other kinds of rotation, like countries and regions.

    And of course there is the full model as well. I have a long term holding period version of the model I don't publish that has a holding period of 3 years.

    There's plenty of room to scale.

  11. That's good to know, as illiquidity is often the quant's enemy #1 :)