tag:blogger.com,1999:blog-261139923818144971.post1794279524728169524..comments2024-03-27T07:58:49.946+00:00Comments on This Blog is Systematic: How fast should we trade?Rob Carverhttp://www.blogger.com/profile/10175885372013572770noreply@blogger.comBlogger21125tag:blogger.com,1999:blog-261139923818144971.post-90940698947255124142022-10-17T10:11:31.011+01:002022-10-17T10:11:31.011+01:00Mostly (1), but I also buffering on the final posi...Mostly (1), but I also buffering on the final position which reduces turnover.Rob Carverhttps://www.blogger.com/profile/10175885372013572770noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-9210283341995669532022-10-17T07:33:20.190+01:002022-10-17T07:33:20.190+01:00Hi Rob, thanks for the post, great as always. I...Hi Rob, thanks for the post, great as always. I'm backtesting your MA crossover rules and I found similar results; the faster crossovers tend to lose money after factoring cost. My question is how do you "slow down" your system? Do you either <br /><br />1) Exclude the fast crossovers rule that are above your speed limit from your final forecast for each asset<br />2) Retain all the rules to get your final forecast, and then use some sort of smoothing to reduce the turnover<br /><br />Thanks! wazewwwhttps://www.blogger.com/profile/08963304023727906040noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-12608714008829727712020-09-30T00:11:53.152+01:002020-09-30T00:11:53.152+01:00As a biginer in systematic trading where can I sta...As a biginer in systematic trading where can I start. amit bhamburehttps://www.blogger.com/profile/15013251312091360540noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-46766903564295007462020-09-14T20:01:51.295+01:002020-09-14T20:01:51.295+01:00ok thanks. I have just bought the second book. Per...ok thanks. I have just bought the second book. Perhaps I should have read it before trying to put something together.Markhttps://www.blogger.com/profile/02096878459039848278noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-52542789260012050062020-09-14T19:57:45.424+01:002020-09-14T19:57:45.424+01:00Yes. I actually address this in my second book, bu...Yes. I actually address this in my second book, but the best solution is probably to replace the bond ETFs with some higher vol products, eg longer duration.Rob Carverhttps://www.blogger.com/profile/10175885372013572770noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-12092808422869999702020-09-14T19:45:53.314+01:002020-09-14T19:45:53.314+01:00yes, the issue is 30% in various bond etfs with ve...yes, the issue is 30% in various bond etfs with very low instrument risk. They are many multiples of leverage as a result causing the overall group of etfs to be 5 x leveraged. Without them leverage drops markedly. Is that the answer?Markhttps://www.blogger.com/profile/02096878459039848278noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-3307301048691809922020-09-14T10:09:29.734+01:002020-09-14T10:09:29.734+01:00What is the natural unlevered risks of the instrum...What is the natural unlevered risks of the instruments you are trading? I'm surprised you need 5 times leverage to hit 57% risk unless you have a lot of low volatility short duration bond ETFs.Rob Carverhttps://www.blogger.com/profile/10175885372013572770noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-48584279215264506902020-09-12T11:30:17.637+01:002020-09-12T11:30:17.637+01:00Rob,
I am trying to implement the full system from...Rob,<br />I am trying to implement the full system from leveraged trading (having done the simple system for some time). I have 24 etf instruments I trade as cfds. They are from many different asset classes. I have calculated the Account Level Risk as 25% and the IDM as 2.3 which gives an instrument level risk of 57.5%. These are from the tables in the book. My understanding is that all positions will be leveraged until their standard deviation % is above 57.5%. This makes it highly unlikely any position will ever be unleveraged unless I go mad and trade crypto. The problem is this usually makes my whole account leverage such that the margin with IG (20% retail) is right on the maximum account value with the full account value as margin, leaving no margin for error(pun intended).Is this just a symptom of using the system as a retail investor(and you intended leverage of 5 x plus to be the norm at an account level) or have I made an error? I don't intend to use the maximum 5 x leverage but would like to know if I am on the right lines in my understanding. Thanks for your books and help.Markhttps://www.blogger.com/profile/02096878459039848278noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-81985703757165175632020-04-24T16:13:00.003+01:002020-04-24T16:13:00.003+01:00Great discussion. It sure sounds like there's...Great discussion. It sure sounds like there's an application of Fourier analysis in there somewhere.nimbusdodgerhttps://www.blogger.com/profile/09041905603837632817noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-61623629519605852372020-04-24T16:08:04.095+01:002020-04-24T16:08:04.095+01:00That makes sense, thanks. The cut off time periods...That makes sense, thanks. The cut off time periods you mention are in line with results I get when looking at optimal delta rebalancing frequency for option trading. No surprise there but it's good to have confirmation.One1hedgehttps://www.blogger.com/profile/08190544206716894643noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-59541315916780314082020-04-24T15:43:19.098+01:002020-04-24T15:43:19.098+01:00Throttling the slower rules wouldn't affect th...Throttling the slower rules wouldn't affect the performance very much, although it would introduce more dispersion in the outcome (since the exact date of rebalancing can affect the p&l, versus it being effectively continous).<br /><br />I also know from past research that using more granular data wouldn't help with the very fast rules. It does seem like the market has more mean reverting behaviour at this frequency. The stylised facts seem to be:<br /><br />Greater than 1 year: Mean reversion (eg equity valuations)<br />~1 month to ~1 year: trend following <br />~1 day to ~1 month: Mean reversion<br />~minutes to day: Trend following<br />sub minute: Mean reversion (market making high frequency)<br /><br />The cut off time periods seem to depend on the asset class, so for example for equity indices the trend following window is narrower than for other asset classes.Rob Carverhttps://www.blogger.com/profile/10175885372013572770noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-26268609357836809712020-04-24T15:33:52.508+01:002020-04-24T15:33:52.508+01:00Thanks for the detailed answers, very clear.
Anoth...Thanks for the detailed answers, very clear.<br />Another question for you: looking at the pre-cost chart comparing LAM to actual SR, the fastest rules seem to decay compared to longer term ones. Do you think it could also come from the granularity of the data you use? If you were to use 1min data to backtest equivalent rules (by that, i mean rebuilding an ewmac 2 days / 8 days from the 1min data, not 2 minutes / 8 minutes), the long term slow rules wouldn't change much with the extra intraday information but the shorter term faster rules would probably rebalance more often and catch more of the intraday trends. Or maybe another way to look at it, if you restricted the slower rules so they can only trade once a week/month (or whatever the equivalent would be of a daily rebalancing for the fastest rules), do you think it would make the performance uniform? <br />Thank youOne1hedgehttps://www.blogger.com/profile/08190544206716894643noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-34605983738434435522020-04-24T09:45:16.370+01:002020-04-24T09:45:16.370+01:00These are all excellent questions.
An important g...These are all excellent questions.<br /><br />An important general point here is that for the sort of trading I am doing getting cost assumptions wrong even by a factor of two has limited effect on my p&l, and almost no effect on the decisions I make in deciding how to trade. It's far better to make a simplifying assumption that means I can use decades of historic data for backtesting.<br /><br />This clearly wouldn't be the case if I was trading faster or in larger size.<br /><br />1. Yes it's true that markets have got cheaper to trade over time. But many markets were also more volatile in the past, though clearly not all. This makes it difficult to apply a blanket rule like 'everything cost more in the past'. Without access to genuine historical slippage data anything you do will be an approximation.<br /><br />This does mean you should be careful; suppose for example you find a fast strategy that worked well in the past but is flat over the last decade or so. Well you know that in the past the costs would have been higher, so really after costs it would have lost money historically. Indeed, this is exactly what happens with faster trend following on equity indices.<br /><br />This suggests using some kind of moving window or exponential mean of returns, although I'd caution that this shouldn't be too short as we will struggle to get any statistically significant results if we start using just a few years for estimation.<br /><br />2. You could do this, and a market where it would make most sense is something like the short bonds or STIR where there was a long period when vol was extremely low (now over?)<br /><br />3. I'd say the real experience is something like this; when the markets get a bit more volatile the inside spread doesn't change but the depth gets much worse. For my relatively small trades that means that trading costs probably don't change much as vol goes up, at least to begin with.<br /><br />When vol gets very high however the spread will increase. This is probably <1% of the time however.<br /><br />Plotting slippage against vol would then probably come out as quadratic(ish) rather than linear; flat to begin with then going up. However it would be a different fit to an institutional sized trader.<br /><br />At some point I ought to do an analysis of my costs data which I have been collecting for several years now to see if I can confirm these hunches.Rob Carverhttps://www.blogger.com/profile/10175885372013572770noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-63474985060248087362020-04-24T09:43:30.159+01:002020-04-24T09:43:30.159+01:00This comment has been removed by the author.Rob Carverhttps://www.blogger.com/profile/10175885372013572770noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-19554116192396690692020-04-23T21:50:58.445+01:002020-04-23T21:50:58.445+01:00Rob,
Thank you for posting all this, very interest...Rob,<br />Thank you for posting all this, very interesting once again. <br />I'd like your opinion on different cases that could impact the model.<br />Just trying to assess if those are worth spending time on.<br /><br />It will be easier with an example and assuming holding costs = 0.<br />For example on SP500, I see in your config a slippage of 0.125 and let's assume the SP500 std dev is 16%.<br /><br />1. Historical half-spreads<br />The 0.125 half spread corresponds to a snapshot of the current market liquidity. If i understand correctly, if SP500 std dev was also 16% back in 1990 (for example), that implies we would assume the half spread to also be 0.125 in 1990 in the backtest. I do not have access to bid-ask data but I'd imagine that spreads were wider then. What is your opinion on this? What do you think about adding a penalty (maybe in %) that increases the further we go back in time?<br /><br />2. Lower volatility<br />Let's assume SP500 vol was 8% at some point in the past (so half the volatility we used when computing the SR costs), the model would imply the half-spread to be 0.125 / 2. But this is not technically possible as the minimum tick size on SP500 is 0.25 so the half-spread could never be less than 0.125. Also, I believe on some futures the current minimum tick size is a fraction of what it was before electronic markets were introduced. Do you think it would make sense to introduce a floor in costs equivalent to that minimum tick size?<br /><br />3. Higher volatility<br />The model assumes slippage moves linearly with volatility. With the most recent increase of volatility, is your live execution in line with this assumption?<br />I know from experience that for option markets the bid-ask usually increases faster than volatility, often in a quadratic way. Interested to know if you observed futures market as being more stable.<br /><br />Thank you in advance!<br />One1HedgeOne1hedgehttps://www.blogger.com/profile/08190544206716894643noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-44481817619675982202020-04-09T16:25:52.580+01:002020-04-09T16:25:52.580+01:00Thank you for the clarification. I am implementing...Thank you for the clarification. I am implementing your formulas and concepts in C#, hence my desire for accuracy.<br /><br />I very much appreciate your efforts in sharing your knowledge through books and this blog. You provide a much needed guide on incorporating risk and cost management into trading.nimbusdodgerhttps://www.blogger.com/profile/09041905603837632817noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-26797105652734432282020-04-09T13:19:41.784+01:002020-04-09T13:19:41.784+01:00This comment has been removed by the author.nimbusdodgerhttps://www.blogger.com/profile/09041905603837632817noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-63379779499346742782020-04-09T10:35:00.226+01:002020-04-09T10:35:00.226+01:00The annualised standard deviation of returns of th...The annualised standard deviation of returns of the instrument.Rob Carverhttps://www.blogger.com/profile/10175885372013572770noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-43869336442659616142020-04-09T00:33:32.597+01:002020-04-09T00:33:32.597+01:00Hi Rob,
Leveraged Trading (Formula 8) uses 'n...Hi Rob,<br /><br />Leveraged Trading (Formula 8) uses 'natural instrument risk'. From the book:<br /><br />Formula 8: Risk-adjusted transaction costs<br />Risk-adjusted cost per transaction = Cost per transaction รท natural instrument risk<br /><br />The example then uses a target portfolio risk of 20% as 'natural instrument risk'. What exactly is natural instrument risk?nimbusdodgerhttps://www.blogger.com/profile/09041905603837632817noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-65576403840612583732020-04-06T09:34:10.650+01:002020-04-06T09:34:10.650+01:00You can go from ATR to standard deviations, which ...You can go from ATR to standard deviations, which you would then use to decide the size of your stop loss gap (X*annual_standard_Deviation) From the second post in this series https://qoppac.blogspot.com/2020/03/how-much-risk-should-we-take.html "If you prefer to measure risk using the well known ATR, then as a rule of thumb multiplying the daily ATR by 14 will give you the annual standard deviation."Rob Carverhttps://www.blogger.com/profile/10175885372013572770noreply@blogger.comtag:blogger.com,1999:blog-261139923818144971.post-6862436831959056852020-04-04T21:02:19.657+01:002020-04-04T21:02:19.657+01:00Do you plan to post the rules for calculating X us...Do you plan to post the rules for calculating X using ATR? Sergey Kushchenkohttps://www.blogger.com/profile/11211081806981013584noreply@blogger.com