Dustin’s Journal

[Anonymous]

A better January… let’s hope the year continues on this trajectory!

MOC/MR Portfolio (US): 4.2%
Nasdaq Rel Mom: 0% (cash)
S&P500 Rel Mom: 0% (cash)

Worth noting that the majority of the returns for the month did not occur when the market was rebounding strongly but rather in the lower volatility flatter days near the end of the month.

[RobGiles]

Dustin Johnson wrote:

And another question that I would be happy to receive thoughts on is how much weight you give to pre-2004 returns when evaluating systems, specifically mean reversion systems. I was listening to the Cesar Alverez interview on BST and it does seem logical that the market has changed so much in terms of quant involvement and market structure that anything before the mid-2000s seems somewhat irrelevant for high-turnover systems like MOC and short term mean reversion.

I am specifically thinking about it because I have been retesting one of the systems that I am trading and, rather than optimizing for a historical period, I optimized for the last 3 years. To test the system robustness, I then extended the OOS period backward. Performance is clearly much better in recent periods (even much better going back to 2010), but as you would expect bigger drawdowns are introduced. However, the big drawdowns are really in 2001-2004 and for no apparent reason (i.e. no market crash etc) and the system does just find during 2008, the 2011 volatility and the 2016 selloff. I am really tempted to trade the more recent parameters given that the system weakness was so long ago.

I would add that I kept the Oct/Nov data as OOS data for testing post-optimization and the drawdown for the updated parameters was 18% vs 14% for the old parameters. So perhaps there is more risk with the new parameters but the returns in recent years seem to more than offset the added risk.

The returns and drawdown tables are attached. The one with large drawdowns in 2004 is the one with parameters updated form the most recent 3 years.

Hi Dustin…where did you end up with the backtesting period for the MR and MOC systems? Did you do as Said suggested and use mid values where robustness was evident?

[Anonymous]

Hi Rob, yes what I have done is used the more recent period for optimization, focusing on selecting robust and stable parameters rather than the peaks. I have also then re-run the final result back to previous periods as well though just to ensure I am not optimizing for a low volatility period that will get slammed if volatility picks up again.

[Anonymous]

Feb Results:

MOC/MR Portfolio (US): 6.4%
Nasdaq Rel Mom: 0% (cash)
S&P500 Rel Mom: 0% (cash)

YTD Results:

MOC/MR Portfolio (US): 10.7%
Nasdaq Rel Mom: 0% (cash)
S&P500 Rel Mom: 0% (cash)

Note that I am now in the market with weekly and monthly rotational systems for the NASDAQ. Makes me a bit nervous buying in here but gotta trust the system.

[TimothyStrickland]

Great results on your MOC/MR strategies mate.

[TimothyStrickland]

Dustin,

Is your MR system taking lots of trades? Mine just started taking a lot of them around the 2nd week in Feb.

[Anonymous]

Thanks for the kind words Tim. Regarding my MR system, I really didn’t take any trades for the first two weeks of the year but since then I have had reasonable exposure… still only keeping up with the market though (which is good considering the market!)

To be a bit more specific, my system allows max 12 positions and YTD has made 46 trades beginning Jan 14th. YTD return is 8.9% for this system. In full disclosure I have relatively little allocated to this system and more of the nominal returns this year of come from MOC systems. They are all performing in the same relative ball park though in terms of returns.

Hope that helps.

[Anonymous]

Mar Results:

MOC/MR Portfolio (US): -3.4%
Nasdaq Monthly Rel Mom: 5.2%
Nasdaq Weekly Rel Mom: 0.8%

YTD Results:

MOC/MR Portfolio (US): 7.0%
Nasdaq Monthly Rel Mom: 5.2%
Nasdaq Weekly Rel Mom: 0.8%

[Anonymous]

A quick update on returns, since I haven’t been posting for a while. I did a major rework of my return calculations to reflect 1) changing capital allocation to different strategies and 2) consolidation of accounts after I changed my IB account. I would also note that my returns look better than they are because there is a survivorship bias in that I have discarded or altered the systems that haven’t been working – all in all, the systems that I have discarded in sum have been breakeven. Many of the discarded systems have been the basis for the systems that I have now landed on but many improvements have been and still are ongoing. I have been experimenting quite a lot and have landed on the systems below.

August has been pretty rough so let’s see where that ends up.

Total Account at July 31: +4.9% (beginning April 1st with new account)

[Anonymous]

Hi all, I realize I haven’t posted performance in a while. Simplifying my numbers, my monthly account returns have been as follows. My systems are ca. 40% rotational, 50% MOC and 10% discretionary and small allocation to new systems in testing. I consolidated my accounts as of May so I am just starting with that month.

May: -4.6%
June: -0.2%
July: +7.8%
Aug: -5.8%
Sept: -7.9%
Oct: -3.5%
Nov: 6.0%

Total YTD (May – Nov): -8.9%

I clearly had a horrible run from Aug to Oct but I took some very large losses being a bit too experimental with leverage and trying to increase exposure. Despite the tough stretch, a number of systems have actually performed quite strongly and I feel pretty comfortable and optimistic about the outlook. I realize its better to just keep a long term view of it and its really not a race… better to make small and steady gains rather than shooting for the moon with crazy volatility.

[Anonymous]

On a separate note, I’m interested to know if anyone is considering, or has already gone down the path of Python for testing and implementation. I picked up the book “Trading Evolved” by Andreas Clenow and he certainly makes interesting claims about the possibilities within Python. I question if longer term we will be required to be using it to remain competitive – a point that Howard Brandy makes a few years ago already.

Also, has anyone done any intraday testing with Amibroker. This is one of the reasons why I am considering looking into Python more. I would like to test a MOC strategy that automatically hedges market exposure but shorting the index in proportion to the long exposure taken on the MOC trade. This way one could really separate the trade alpha with the beta exposure. This seems really tough in Amibroker for a number of reasons, first due to the use of databases (i.e. Norgate which has historical constituents and any other intraday data feed which doesn’t have historical constituents) but also due to the implementation alternatives even if one could test the idea accurately.

[Anonymous]

Always nice to have a great month and then get smashed the first trading day of the following month… :cheer: but seriously, it’s (hopefully) a good sign that it now just seems part of the process. I think it was this time last year when I was sweating about a 2% drawdown.

[SaidBitar]

Dustin Johnson wrote:

On a separate note, I’m interested to know if anyone is considering, or has already gone down the path of Python for testing and implementation. I picked up the book “Trading Evolved” by Andreas Clenow and he certainly makes interesting claims about the possibilities within Python. I question if longer term we will be required to be using it to remain competitive – a point that Howard Brandy makes a few years ago already.

Also, has anyone done any intraday testing with Amibroker. This is one of the reasons why I am considering looking into Python more. I would like to test a MOC strategy that automatically hedges market exposure but shorting the index in proportion to the long exposure taken on the MOC trade. This way one could really separate the trade alpha with the beta exposure. This seems really tough in Amibroker for a number of reasons, first due to the use of databases (i.e. Norgate which has historical constituents and any other intraday data feed which doesn’t have historical constituents) but also due to the implementation alternatives even if one could test the idea accurately.

Few months back i had some ideas that should be tested on intraday data, so i tested using amibroker.
the ideas where having multiple exits during the day such as profit taking and if not touched MOC, but after the testing the results were not better that daily data with MOC so i stopped.
this is regarding to intraday data.
Regarding python and Andreas book they are nice but using python for backtesting is not different from Amibroker, Ok on the surface it seems to give you more control but if you are using CBT in Amibroker will be the same result with less effort.

[Anonymous]

Thanks Said. Regarding intra-day testing, that is interesting and super helpful. How did you combine intraday data with the norgate historical constituents watchlists? I think you are probably exactly right it that, like most ideas, it seems like it would be interesting and then really doesn’t add any thing.

Regarding python, I think if you are doing the exact same type of trading (i.e. indicator-based, price/volume-based, end of day data etc.) then amibroker will definitely be easier. I wonder though if you can use a machine learning approach to help identify tradable patterns though. Even if you then implemented in Amibroker. At the end of the day, you are probably right again in that it seems like it would be advantageous, but prove to be difficult, time-consuming and really no different than amibroker. Can you code in Python? For me that is the first big hurdle. But if you tell me that you can already code in python and don’t see it adding any value then it is a complete waste of time to try that route.

[ScottMcNab]

Dustin Johnson wrote:

Also, has anyone done any intraday testing with Amibroker. This is one of the reasons why I am considering looking into Python more. I would like to test a MOC strategy that automatically hedges market exposure but shorting the index in proportion to the long exposure taken on the MOC trade.

Another option may be to use futures to short the index ?…develop a system that goes short on breakout to downside…these are the days MOC would be loaded up…can test using tradestation intra-day data but would need to learn easylanguange

[Author]

Posts