Daily Fibonacci Signals

[snailbeard]

We can now load and fingerprint Zorro and TickStory CSV files, perhaps I should add Empty4 hst files for completeness?

So just as you expected the tickstory data is so much better than Zorro's data:

In a previous post is the Zorro report,
The TickStory summary:

Code: Select all

 
 sumMinutes       : 373018
 sumDuplicateMins : 0
 sumRepeatMins    : 0
 sumMissingMins   : 528
 sumSpuriousMins  : 0
 sumExtraMins     : 0

So the only issue is missing minutes, but I need to check the details as 528 is an number which suggests a systematic reason rather than random data loss.

The the finger-print values should correspond with correlated assets:

Code: Select all

Major M1 Bar moves:
Time: 2014.09.07 22:00:00 Abs%:  1.00 Price: 1.61615 Change: 0.01610 dir: -1, Prev Time: 2014.09.05 21:59:00 Prev Price: 1.63241
Time: 2014.01.17 09:30:00 Abs%:  0.47 Price: 1.64225 Change: 0.00795 dir: +1, Prev Time: 2014.01.17 09:29:00 Prev Price: 1.63430
Time: 2014.01.10 13:30:00 Abs%:  0.33 Price: 1.64489 Change: 0.00408 dir: +1, Prev Time: 2014.01.10 13:29:00 Prev Price: 1.63808
Time: 2014.01.10 09:30:00 Abs%:  0.22 Price: 1.64138 Change: 0.00346 dir: -1, Prev Time: 2014.01.10 09:29:00 Prev Price: 1.64485
Time: 2014.01.02 14:59:00 Abs%:  0.13 Price: 1.64230 Change: 0.00186 dir: -1, Prev Time: 2014.01.02 14:58:00 Prev Price: 1.64422
Time: 2014.01.02 09:28:00 Abs%:  0.11 Price: 1.65400 Change: 0.00164 dir: -1, Prev Time: 2014.01.02 09:27:00 Prev Price: 1.65564
Time: 2014.01.02 09:07:00 Abs%:  0.10 Price: 1.65834 Change: 0.00147 dir: +1, Prev Time: 2014.01.02 09:06:00 Prev Price: 1.65687

[/size]

The above method is a first attempt from which to think about a better method of making sure a basket of assets line up. Once we have loaded a basket of currencies we cannot compare prices directly, we are interested the relative change.

The 4 largest events of the above correspond almost exactly with the dirty Zorro data, whereas as one event from each finger-print does not correspond and we can never expect a one to one match but an 80% match should be sufficient.

[Radar]

Hey SnailBeard,

If you export some of the DukasCopy tick data to a CSV, you'll find buy and sell volumes for each tick...
DC may be large enough to use that info for Volume Spread Analysis... Might be worth a look...

Have fun!
Radar =8^)

[snailbeard]

Hey SnailBeard,

If you export some of the DukasCopy tick data to a CSV, you'll find buy and sell volumes for each tick...
DC may be large enough to use that info for Volume Spread Analysis... Might be worth a look...

Have fun!
Radar =8^)

That's an interesting topic which some people have studied in depth, perhaps it applies more to shares than currencies?

[Radar]

Hey Snailbeard,

That's an interesting topic which some people have studied in depth, perhaps it applies more to shares than currencies?

Yeah, the debate on the validity of VSA in forex has been raging for quite a while, since Empty4's volumes are broker specific, and only shows the number of transactions without any indication of their size.
Some people swear by it, whilst others swear at it. Me, I'm having a hard enough time learning the terminology, let alone learning how to apply the rules

Have fun!
Radar =8^)

edit... I shouldn't post when I'm half asleep... Just realised that when I exported DC tick data to CSV, it showed transaction data per tick. I'd have to learn Java to be able to make any use of it, though.

[snailbeard]

When currencies are bouncing around in no particular direction it can be useful to look at trading something else.

An old friend of mine is Riotinto, always doing something and also a good finger burner. I used to get burnt by the morning gaps, yes you've been there too

Now I wonder if I might ride some of those gaps...
So how would that work?

Firstly, it means entering a trade the day before, perhaps near the end of the day.
Sometimes you can place both buy and sell at the same time and win both if you exit at the right time.
Has anyone here tried it and achieved consistent results?
However, I was thinking mainly of a single trade when there seems to be momentum in a particular direction.

I have done some manual testing based on a combination of momentum and volume indicators.
However, I'd like to test a range of shares with some historical data.

The first trap is that back-testing with some indicators can give spurious values when we test over a narrow date range particularly OBV, MFI, and A/D. I was experimenting we these to see if they helped improve a guess at following days price movement.

Although, I am getting some good results manually, I might be subconsciously looking at other patterns in the chart.

The other issue is 'lot size annihilation' or 'death by lot size' - due to the quirky nature of lot sizes for different assets: if we are not careful, a bit of finger trouble could turn a 1% risk into 100% risk

A lot size of 10 for shares does very little, whereas a lot size of 1.0 for Brent could be a disaster.
One risk calculator says I should use 10,000 for BP, 2.5 for WTI, and, 0.01 for Brent .
My home-brew forex risk calculator suggests 1000.0 for shares, and 0.06 for WTI, and 55 for Brent


So it seems I should stick to currencies and shares, and avoid the commodities for time being until I have a consistent risk-calculator.

Here is a bit of code that enables us to have indicators fully loaded but only runs our own code between specified date:

Code: Select all

int start() {

   static bool st_bSartedTesting = false;
   static bool st_bReportedNotYet = false;
   static bool st_bReportedStoppedTesting = false;
   
   datetime dtFebEnd = D'2016.02.29 00:00';
   datetime dtTestFrom = dtFebEnd;
   datetime dtTestUntil = D'2016.03.03 00:00';
     
     string strTestStart = TimeToStr( dtTestFrom, TIME_DATE); 
     string strTestEnd = TimeToStr( dtTestUntil, TIME_DATE); 
     
     if( IsTesting() )
     {
         if( TimeCurrent() >= dtTestUntil)
         {
            if( st_bReportedStoppedTesting == false )
            {
               Print( "Testing stopped at: ", strTestEnd );
               st_bReportedStoppedTesting = true;
            }
            return(0);
          }
         
           if( TimeCurrent() >= dtTestFrom )
           {
               if( st_bSartedTesting == false )
               {
                  st_bSartedTesting = true;
                  Print( "Testing starts from NOW!" );
                  Print( "Testing ends at: ", strTestEnd );
               }
           } else {
        
               if( st_bReportedNotYet == false )
               {
                  Print( " Skipping start() until: ", strTestStart);
                  st_bReportedNotYet = true; 
               }
               return(0);
          }
     }

[/size]

This avoids time spent wondering why the on-screen indicator values don't match the numbers coming out of crappyTester, and reduces the amount of data to wade through in the log file.

Although, I had some success with the editors debugger when all the code is in one file, breakpoints have no effect when they are in header files

Next time I'll post some Riotinto charts with interesting volume indicators.

[snailbeard]

Although, this talks about gapping, sometimes the day opens without a big move, which is fine if the take profit is based on ATR. Mostly a target of (D1 Atr / 2) seems a good option.

The following chart covers about two weeks of trading, a large tick ✔ ( not a price tick! ) means the price moved enough in the direction that was predicted near the end of the previous day. The results are interesting but not statistically meaningful:

riotinto-guess-rsults-2weeks-Mar2016.png

There is a strong possibility that this short test is a result of curve fitting. Experienced traders know that indicators are only useful some of the time and I've used RSI overbought and oversold to avoid some losers:

riotinto-oversold-03Mar2016.png

riotinto-overbought-7Mar2016.png

Often a trend continues, even when indicators show overbought or oversold, it is risky to take it as an opportunity to reverse without additional information. For example the pattern of the last few bars are often associated with a reversal.

Some traders use volume and price divergence to predict reversals. In the following chart we see that both MFI and RSI do not make higher-highs with the price. The above test does not include divergence.

riotinto-3D1-H1-MfiDiv-RsiOverbought-7Mar2016.png

The next step is to rerun the predictor on different data to check if it is better than just curve fitting...

[snailbeard]

The good news is that "End-of-Day Guess Tomorrow" (EODGT) works on some assets including spot gold, I am going to run this EA live next week on a small number of assets with a reduced risk, to see what happens in the real world.

One interesting result comes from testing on BARC which produces more losses than wins even though the guesses are consistent with the longer term. If we consider each asset to have its own personality, then we can stop trying to use the same algorithm on different assets, although we can find similar personalities in different asset classes, so if we look at EURCHF and BARC side by side we might expect the same result from them.

What is the correct algorithm to use on EURCHF and BARC?

In the following chart we can see that the daily bi-directional swings prevent us from using the EODGT approach:

#Barc-EoDGN-losses-due-to-swings-Mar2016.png

Clearly, the only way to trade the above is to enter in the direction of the guess when the price has moved "far enough" in the opposite direction. This might work some of the time if we can determine a reasonable value for "far enough".

[snailbeard]

We normally read about Fibonacci in the context of price movements. Googling Fibonacci with other keywords related to volume or accumulation/distribution did not return any hits in the first couple of pages.

So does it make any sense to use Fibonacci levels on an A/D line?

Although, I have not done enough research to answer this question or thought about the relationship of the A/D values to the price values (some indicators derived from price could be expected to have coincident Fibonacci levels). In the following chart the price high/low and retracements do not correspond directly to the A/D high/low/retracements but they both suggest closing at or under resistance.

barc-fib-pattern-on-AD-line-01Feb2016.png

Does anyone find this to be an interesting and possibly useful observation?
Or have you already explored this in the past?

Is this an artefact of the formula for calculating A/D?

[snailbeard]

In this second chart the Fibonacci levels are less helpful, but the MFI diverges from both price and A/D, hinting that there might be buying pressure, which turns into an up-gap the following morning:

BARC-Price-AD-fall-while-MFI-rises-11Feb2016.png

My attempts at using yesterday's guess after a morning reverse gap have returned some successes but a profit ratio of 1.1 is not very interesting.

The previous two charts hint at the direction of the next move and I might do better to spend time improving EODGT than looking for new ways to trade assets with bidirectional swings.

[snailbeard]

In this final chart, we see a good example of both AD and MFI going in the opposite direction of price, it hints that the next move will be up and the next day opens with a gap up. The Fibonacci lines are not required unless they can help us with a target for take profit. Since, I generally, use a simple derivative of ATR to select profit, using Fibonacci levels, may or may not be appropriate and would require more logic and testing to find out.

BARC-M15-Price-MFI-AD-vol-divergence-19Feb2016.png

Divergence doesn't guarantee a reversal, but two volume indicators telling the same story is better than just one!