CL always-in

This is something I've always been interested in testing but I simply don't have the wherewithal to do it. Could you perhaps post a chart with trades or what you might consider candidate setups overlaid on it so I can have a better feel for what your system is looking for. This makes sense in theory but it seems like it would only be efficient on very large time frames.

Have you tried this with any other instruments like NG or any of the currencies?
 
Quote from riskaddict:

This makes sense in theory but it seems like it would only be efficient on very large time frames.
More...

My guess is it`s on the contrary.You can`t get enough signals from the higher frames,such as DT,within RTH.So my view, it uses the finest ones.
 
Quote from Quiet1:

hi - thanks for sharing. this is quite interesting.

if it's not giving too much away how do you achieve the orthogonality of your patterns on each tier?

how do you handle conflicting patterns between tiers (ie if tier 1 says long, tier 2 says nothing, tier 3 says short and tier 4 says X). Smallest/shortest timeframe tier always wins but reverts to higher timeframe position when closed?

q1
More...

Orthogonality is ensured by design, ie. I select patterns in each tier that I know cannot overlap (for example, I have 1 tier dedicated to the various ToD bias, 1 tier dedicated to the total trend-size relationship between last trend and the prior one).

There are a few patterns that could overlap in one tier, but I resolved that by ranking them.

In all cases, the system uses the 1st pattern found as a match to the current set of circumstances, so there is no possible conflict in real-time.

I have assigned patterns to tiers depending of their overall quality, which I measure using a Sharpe-like metric (SQN if you are familiar with Van Tharp, Information Ratio for others, anyway it is avg(x)/st-dev(x) where x is the outcome of the pattern).
 
Quote from riskaddict:

This is something I've always been interested in testing but I simply don't have the wherewithal to do it. Could you perhaps post a chart with trades or what you might consider candidate setups overlaid on it so I can have a better feel for what your system is looking for. This makes sense in theory but it seems like it would only be efficient on very large time frames.

Have you tried this with any other instruments like NG or any of the currencies?
More...

The system process intraday trends, about 12,000 over 2007-2012, which is likely an average of 6 per day (more in 2008 & 2011). See the attached trend-size distribution.

I have started a similar approach using a simplified trend-detection mechanism on ES & 6E, but didn't invest enough R&D time on these markets yet.

I won't post charts - in general, patterns are too large to be visible on a single screen size (I use 100-volume to create the intraday Highs/Lows, from which the trend-detection mechanism decides the trend-changes), and what a particular pattern look like doesn't matter - I found all of them through statistical analysis of a number of datapoints recorded with each trend, the price patterns are based on classic relationships between trends (eg, last trend size / prior trend size).
 

Attachments

Quote from cdcaveman:

how are you concatenating contracts together?
More...

For backtesting purpose: NinjaTrader merge-backadjusted, with love and care for the rollover database.

For live-trading purpose: I do that using my own software, since Ninja can't do it from 6pm to mid-night on rollover day. I have the "old" contract in one TF, the "new" contract in another TF, I measure the price difference between the two using the daily pivot calculated for both (H+L+C), then I forward-adjust the prices I get for the new contract (so that they match the prices already in for the old contract) in a proprietary framework which has the same interface as the native Ninja interface (High[] / Low[] / ...), so that forward-adjustment mechanism is 100% transparent to the strategy.

I have about 3500 lines dedicated to that mechanism, I still had to restructure my code on top of that to workaround yet another bug in Ninja's framework (essentially, going from 1 strat + 2 indicators for pivots & trend, to 1 strat + 2 proprietary software components for pivots & trend)
 
Quote from dom993:

(I use 100-volume to create the intraday Highs/Lows, from which the trend-detection mechanism decides the trend-changes)
More...

Why don`t you try a 1000V one,for e.g.,and bust all caps in that cartridge clip at once?(while JH is messing around with the objects)
 
Results for the week ending June, 21st:

- 10 wins / 4 losses ; net +5270.

The system made a new P&L peak on Thursday (+15715 on my trading account), then got caught long a good part of Friday, giving back a good chunck, before getting 2/3 of that back in 2 trading-decisions 10min apart in the close. Drawdown on Realized P&L as of the last trade closed: -425.

Unfortunately, things were not smooth on the operations side. I had Thursday a discrepancy between my back-up PC reversing to long, and the VPS staying short. A deep look at both charts revealed some minor differences in data, I reloaded manually on both sides, no change :( ... I did more investigation, pin-pointing some missing data on the VPS between 3am-3:37am, then I thought of using the Historical Data Manager to reload, instead of CTRL+SHIFT+R ... and it worked (I knew it immediately, as the strategy reversed to long right there and then, as it is supposed to do). I then spent a lot of time, during the day, overnight (until 4am) and today, slowly narrowing the issue, which turns out to be affecting CTRL+SHIFT+R (and its underlying API, which I use in my own software) when there is a strategy active on a chart :( I am in contact with NinjaSupport, hopefully this time their R&D team will be more cooperative than in the past, else it is going to be very difficult for me to workaround this one.
 
Results for the week ending June, 28st:

- 12 wins / 5 losses ; net + 2005

The system had a long winning streak Monday to Wednesday noon ( 9 wins / 1 loss ; net +2550), making a new P&L peak in my account at +17,840. Then it got caught short until Thursday evening, erasing -1415 in 1 loss. Overnight & Friday was hit & miss (3 win / 3 loss ; net +850 ; unrealized P&L -300).
Current drawdown from P&L peak: -545 (realized P&L only).

I spent Saturday evening then all day Sunday (really, from 5am to 5pm) working on that historical data reload issue, first narrowing the issue to what it is, and this is:
1. Ninja CTRL+SHIFT+R (and underlying API) does NOT reload the chart data after successfully downloading the required historical data, when 2+ timeframes are in use on that chart (either directly on the chart, or through a strategy - my situation).
2. Ninja HistoricalDataManager / ReloadAll doesn't do any better job in the same situation.
3. Ninja HistoricalDataManager / Download, luckily, doesn't have that problem (but it triggers a f***g modal confirmation box if any strat is running on minute-bars or tick-bars, luckily again for me, strategies running on volume/range/second-bars are ignored).
I duly documented those findings (1 & 2) for the Ninja support team, and started right away working on using the HistoricalDataManager / Download through Reflection, to at least have a workaround in place while Ninja fixes issue #1, or gives me a better workaround (why do I have that old Supertramp tune playing in my head right now : "Dreamer, you're nothing but a dreamer, ..."). By Sunday 5pm I had a workaround waiting for live market for final testing, and by Monday evening that workaround was deployed on my VPS. Of course, it is Friday evening, and I haven't heard back from Ninja support (yet).

I had another operation issue this week, no harm but it triggered me to change one of Ninja's Data options to "Save chart data as historical" on the VPS.

As it is the end of the quarter, here are the stats straight from my trading spreadsheet, since I started trading that system live on March-19.

(really, it doesn't make justice to the large drawdown in April ... -8,270 on April 23 - from the 1st P&L peak +955 on March 22)
 

Attachments

You must log in or register to reply here.
Back
Top