Denmark
Age 51
Trading strategy
Pair trading strategy, statistical arbitrage, developing math based algo's.

Machine learning based algo's.
Follow me on

Disclaimer

The Saxo Bank Group entities each provide execution-only service and access to Tradingfloor.com permitting a person to view and/or use content available on or via the website is not intended to and does not change or expand on this. Such access and use are at all times subject to (i) The Terms of Use; (ii) Full Disclaimer; (iii) The Risk Warning; (iv) the Rules of Engagement and (v) Notices applying to Tradingfloor.com and/or its content in addition (where relevant) to the terms governing the use of hyperlinks on the website of a member of the Saxo Bank Group by which access to Tradingfloor.com is gained. Such content is therefore provided as no more than information. In particular no advice is intended to be provided or to be relied on as provided nor endorsed by any Saxo Bank Group entity; nor is it to be construed as solicitation or an incentive provided to subscribe for or sell or purchase any financial instrument. All trading or investments you make must be pursuant to your own unprompted and informed self-directed decision. As such no Saxo Bank Group entity will have or be liable for any losses that you may sustain as a result of any investment decision made in reliance on information which is available on Tradingfloor.com or as a result of the use of the Tradingfloor.com. Orders given and trades effected are deemed intended to be given or effected for the account of the customer with the Saxo Bank Group entity operating in the jurisdiction in which the customer resides and/or with whom the customer opened and maintains his/her trading account. When trading through Tradingfloor.com your contracting Saxo Bank Group entity will be the counterparty to any trading entered into by you. Tradingfloor.com does not contain (and should not be construed as containing) financial, investment, tax or trading advice or advice of any sort offered, recommended or endorsed by Saxo Bank Group and should not be construed as a record of ourtrading prices, or as an offer, incentive or solicitation for the subscription, sale or purchase in any financial instrument. To the extent that any content is construed as investment research, you must note and accept that the content was not intended to and has not been prepared in accordance with legal requirements designed to promote the independence of investment research and as such, would be considered as a marketing communication under relevant laws. Please read our disclaimers:
- Notification on Non-Independent Invetment Research
- Full disclaimer

139 Followers
Write a Squawk to pairstrader1
  • Squawk / 02 August 2017 at 19:53 GMT
    -
    Denmark
    Algo update.

    In my search for features to use with my algo and my deep learning neural network filter, I tried many things. Classic SMA, MACD, RSI etc. None of them work with reliable results, best case slightly better than 50:50.

    However, a combination of time-of-day parameters:

    day-of-month
    day-of-week
    hour-of-day
    minute-of-hour

    Rationale for the time-filter is that trades sometimes get caught on the wrong side just after econ. news has been released

    Above, not surprisingly, do add value to the DNN filter. In fact, it increases the prediction of the signals being profitable from around 66% to around 75%

    Applied to my algo overview page, all signals last profit/loss ratio went from 72% to 80%, meaning my 100 algos running now, 80% is now in profit of the last trade whereas 72% without the DNN filter.

    I turned on the algo on friday around noon, and until now it has booked 8%. Most of it is probably luck, but i will keep it running for the coming weeks to validate the results
    Read the Squawk
    2y
    fxtime fxtime
    day of month/day of week/hour of day/ minute of hour and TICK totally agree with. Time is a crux parameter IMHO for risk and trade entry.
    2y
    pairstrader1 pairstrader1
    Good point in looking at trading days instead of day in month. It must the cleansed from non-trading days like weekends, since weekends has this weird tendency...
    2y
    pairstrader1 pairstrader1
    Btw, what to you mean about tick-bias?
  • Squawk / 02 August 2017 at 6:29 GMT
    -
    Denmark
    Algo update.

    Attached is a simulation for the accumulated returns over a series of trading signals, one presented in raw form and the other has been filtered by a deep-neural-network. Guess which is which.
    Read the Squawk
  • Squawk / 28 July 2017 at 15:25 GMT
    -
    Denmark
    Algo update.

    It has been a while since my last post. During this time I have been working hard on getting my FX pair trading algo system to utilize machine learning to get even better trading results.

    It can be boiled down to:
    * Each algo's parameters are optimized, to generate best possible sharpe ratio, on in-sample data and verified on out-of-sample data.
    * All algo's signals are processed by a deep neural network (NN), where metrics are used to predict the probability of the profitability of the signal. These metrics are variant computations of ROC, BOLLINGER etc. generated from data per leg in the pairtrade. All in all around 40 features are used.
    * Several algos are combined in a portfolio, where weighting in the portfolio is determined by maximizing the sharpe ratio of the total portfolio.

    Performance of the portfolio, where all signals are used in "raw" for has a sharpe ratio around 5.5. Processing signals with the NN, increases it to sharpe to 8.5.

    (red=out-of-sample)
    Read the Squawk
  • Squawk / 05 April 2017 at 20:12 GMT
    -
    Denmark
    FED minutes, and USD and the SP500 flushes yet again.

    I am beginning to wonder if FED is doing this on purpose.
    Read the Squawk
    2y
    fxtime fxtime
    LOL....we all think that way :-)
  • Squawk / 23 March 2017 at 10:02 GMT
    -
    Denmark
    Great overview article on AI/ML

    https://medium.com/@NathanBenaich/6-areas-of-artificial-intelligence-to-watch-closely-673d590aa8aa#.d71i0bua5

    If you plan to long term invest in this area, many of the leading companies are listed in the article. Only problem is, most of them are not public and others are not just doing AI/ML.

    On investment strategy could be: Buy a basket of them all, wait 2-3 years and see what happens.

    * Google (GOOGL)
    * Microsoft (MSFT)
    * Nvidia (NVDA)
    * Mobileye (MBLY) recently acquired by Intel (INTL)
    * Facebook (FB)
    * Apple (AAPL)
    * ATI (not mentioned in the article, but they are players too)
    * Amazon (AMZN), only providing platform for ML computing
    Read the Squawk
    2y
    fxtime fxtime
    You don't mention Elon Musk ? Whilst he expresses concerns about AI and its application I suspect Tesla should be applied to the above (regardless of Elon...
    2y
    fxtime fxtime
    It is a good subject matter Pairstrader and every plaudit to you for bringing it to the fore.....however I suspect it will be a subject matter that...
    2y
    pairstrader1 pairstrader1
    Well, yes, i kinda forgot Tesla. They are kind of being a frontrunner in autonomous driving, at least one of the more visible in the field. Their...
  • 2y
    Michael O'Neill Michael O'Neill
    Thanks for that. You are right. The comment is out of context.
    05 January
    modeons modeons
    The overreaction and finding the impressive thoughts about distorts trading the great college methods with us. The great https://www.australian-writings.org/ sessions and implementing the excited started higher spiking...
    09 March
    eddie007 eddie007
    I found this tutorial so useful because here i get to know about how to remove search records of http://deletebinghistory.com bing search history delete for this search...
  • Squawk / 15 March 2017 at 20:31 GMT
    -
    Denmark
    Every time I hear the media whine about BREXIT and how bad things are going to be, why is that I get to think about a quote from the movie called Margin Call.

    >> If you're first out the door, that's not called panicking
    Read the Squawk
  • Squawk / 15 March 2017 at 20:22 GMT
    -
    Denmark
    EUR is barely moving, steep rise in EURUSD is all about the USD.

    AUD and NZD was better longs against the USD, not the EUR
    Read the Squawk
  • 2y
    fxtime fxtime
    lol
  • Squawk / 14 March 2017 at 13:23 GMT
    -
    Denmark
    Algo update.

    It's been a while since my last post. I have been busy looking into techniques to strengthen machine learning algos when data set are skewed (e.g. few positive classes, many negative), which is the case on trading signals.

    This brought me into the areas of sub-sampling, boosting and ensemble learners. Very technical stuff, very interesting ;-)

    Ensemble leanings is about combining several ML algos'. Lets say that you have 3 ML algos, where two are classifying the signal as 'buy' and one is saying 'dont know', then the consensus is a buy signal. Trick is to give the three algos different sub-sets of the larger dataset.

    Sub-sampling is picking the sub-sets randomly so that the each training set for each of the ML algos has an even amount of 'buy" and "dont know"'s. This can result is better prediction instead of having one ML algo that tries to learn it all.

    Further, splitting it up into multiple ML algos, also makes it possible to run in parallel.

    ... to be continued.
    Read the Squawk
    2y
    pairstrader1 pairstrader1
    about classification. I am looking into training the ML algos in one- vs-others strategy e.g. "buy" vs. "dont know" or ">2% gain over one week" vs. "dont...
    2y
    fxtime fxtime
    I would be intrigued to learn of your outcomes with this :-)
Check your inbox for a mail from us to fully activate your profile. No mail? Have us re-send your verification mail