I don't understand

Discussion in 'Automated Trading' started by BuySellSideTrader2020, Nov 13, 2018.

  1. Yes
     
    #31     Nov 14, 2018
  2. gaussian

    gaussian

    You're correct - I addressed that in an edit. The funny thing about sharpe ratios is they can be gamed by increasing the tick frequency.

    Regarding your second statement I would tend to disagree. Unless this guy is a patent genius his problem is probably simpler than fills. Market microstructure is incredibly difficult to model. I haven't done it myself (books have sufficiently dissuaded me from trying) but it seems that the signal to noise ratio is so high if you're not making markets you're not making money.
     
    #32     Nov 14, 2018
  3. sle

    sle

    I must be misreading. Sharpe ratios are in general presented in a normalized (usually annualized) form. Tick frequency should not matter in that case, right?

    For starters, what is your execution model - do you take or provide liquidity? If mixed, what's the proportion? When you back-tested, what was the pnl/trade value? On your back-test, what was your fill assumption? if passive I'd use trade through, and in general it's conservative to assume taking.
     
    Last edited: Nov 14, 2018
    #33     Nov 14, 2018
  4. gaussian

    gaussian

    Sharpe Ratios can be gamed because the average profit of a trade scales with the number of trades while the volatility scales at the square root of the time. In other words by trading at a higher frequency you can increase your sharpe ratio almost indefinitely.
     
    #34     Nov 14, 2018
    tommcginnis likes this.
  5. sle

    sle

    Oh, you are talking about in back-test? Right, as you move your trades into higher frequencies, your Sharpe increases at the cost of drastic reduction of profit/trade. In real life, if anything, you'll find that your trading costs will start increasing much faster than trading profits unless you have the right technology.
     
    #35     Nov 14, 2018
    gaussian likes this.
  6. sle

    sle

    Futures are usually single-listed so the broker is showing you the actual order book from the exchange. The broker is not making a market but rather passing the orders to the exchange. However, if your broker (and/or your personal technology) suck, you might be getting picked off by the faster players.
     
    #36     Nov 14, 2018
  7. qlai

    qlai

    #37     Nov 14, 2018
    tommcginnis likes this.
  8. sle

    sle

    It's pretty much a given that if you are working on anything latency sensitive, you should write your own backtesting engine. If you are assuming partially passive execution, it get's really hairy and you can't really trust the results that much.
     
    #38     Nov 15, 2018
    tommcginnis and qlai like this.
  9. volpri

    volpri

    You have add two ticks on price movement. One tick on each side. Sims generally don’t so they aren’t realistic.
     
    #39     Nov 15, 2018
    schweiz likes this.
  10. schweiz

    schweiz

    Exactly. And if then the system is profitable in backtesting, it will be probably profitable in real trading too.

    Don't trade in very small timeframes and not too frequently. Small timeframes have only small profit potential while the risk is big. Bigger timeframes have more potential and almost the same risk as in the small timeframes. So the potential for profit increases while the risk stays almost the same.

    The difference between the high and the low is much bigger in 60 minute charts then in 1 minute charts.
    And to make profit that's what you need: a big difference between the high and the low.

    In general HFT trading in not for 20 year old boys unless your are a genius.
     
    #40     Nov 15, 2018
    tommcginnis likes this.