Why exactly do EMA crossover systems not work? | Page 27 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Help. EMA Crossover with different EMA cross for exit 0 replies

EMA crossover (ema 14 & 34) 4 replies

Please help code an EMA Crossover 12ema x 24 ema cross 1 reply

Systems, Systems and Systems... 1 reply

EMA Crossover Systems 1 reply

Attachments: Why exactly do EMA crossover systems not work?
Exit Attachments

Why exactly do EMA crossover systems not work?

  • #521
  • Jan 14, 2016 2:54am Jan 14, 2016 2:54am
To me the reason is simple: this method is simply nonsensical! The line of the MA is an estimator of the equilibrum price. This means it is the best estimation of the fair value within your time horizon of trading. When you buy at the price just above the line you're buying a bit more expensive than the fair value. The job of a speculator is to find mispricing not to buy and sell at fair value. Money is made only if fair value shifts significantly. As an investor (not short term trader), if your fundamental analysis makes you confident the fair value will increase there is anyway no reason to wait up for the price to cross the line if you can buy cheaper right now.
No greed. No fear. Just maths.
  • #522
  • Jan 14, 2016 6:26am Jan 14, 2016 6:26am
Quoting PipMeUp
To me the reason is simple: this method is simply nonsensical! The line of the MA is an estimator of the equilibrum price. This means it is the best estimation of the fair value within your time horizon of trading. When you buy at the price just above the line you're buying a bit more expensive than the fair value. The job of a speculator is to find mispricing not to buy and sell at fair value. Money is made only if fair value shifts significantly. As an investor (not short term trader), if your fundamental analysis makes you confident the fair...
Good points. I just realized that I posted the 2nd reply to this thread in 2007 to this very old thread! And almost as surprising is that my reply way back then was somewhat sensible. In reply to that post and in reply to your post I decided to test whether the MA is actually representative of "value" and whether price tends to return to these historical value levels and at what frequency for a reasonable forward looking window of 5 times the moving average length. Other forward looking windows could also be used but 5*MALength seemed reasonable to me.

It should be noted that the tests below aren't a trading strategy but if it were it would be a very inefficient grid trader that entered every time the Close price was away from (above or below) the MA and exited either on convergence to the original MA level, or at 5 * the MA length (presumably a time-based stop).

This was EUR_USD 15 minute data in case anyone wants to reproduce this or continue with this research.
Inserted Code
      MA Length Convergence %           Pnl
 [1,]         5     0.8713410 -4.639465e-06
 [2,]        10     0.8632604  8.312687e-08
 [3,]        15     0.8669159  2.807925e-05
 [4,]        20     0.8689755  4.349835e-05
 [5,]        30     0.8621971  7.414896e-05
 [6,]        40     0.8654235  1.170542e-04
 [7,]        50     0.8597992  1.319658e-04
 [8,]        75     0.8587644  2.341944e-04
 [9,]       100     0.8552978  2.016942e-04
[10,]       150     0.8497460  2.348945e-04
[11,]       200     0.8450421  4.149805e-04
[12,]       250     0.8412245  2.427027e-04
[13,]       300     0.8423665  3.337410e-04
[14,]       350     0.8417359  3.912159e-04
[15,]       400     0.8343330  2.208804e-04
[16,]       450     0.8263061  2.212193e-05
[17,]       500     0.8288002  1.050849e-04


Predictably the shorter the MA length, the higher the probability that the price converges to the historical MA, but also with lower profit potential. Also there is more profit opportunity in general in convergence back to longer term MAs, at least up to about 200-350 in this example, and then the profit potential starts to diminish on longer MA lengths.

In contrast to the statistics posted above, a typical trading strategy might only enter when price was significantly away from "value" such as 2+ deviations away, and then play the convergence to the current MA level, rather than the historical MA level. This would lead to larger profit potential but possibly a lower hit rate. I'll leave this future analysis as an exercise for the next person as this was just intended as a quick proof of concept.

R code to reproduce these statistics above is attached.
Attached File(s)
File Type: txt Is_MA_Value.txt   2 KB | 448 downloads


I ran the test again using a future window of only 2 * the MA length to get an idea of how quickly price tends to converge to these historical means, and whether there is any profit potential for quicker reversion windows. Here are the results:

Inserted Code
      MA Length Convergence %           Pnl
 [1,]         5     0.7956243  8.470802e-07
 [2,]        10     0.7840526  5.187528e-06
 [3,]        15     0.7767130 -1.442262e-05
 [4,]        20     0.7695328 -1.715492e-05
 [5,]        30     0.7717281  1.745184e-05
 [6,]        40     0.7787759  7.042644e-05
 [7,]        50     0.7829016  1.171658e-04
 [8,]        75     0.7708350  1.990734e-04
 [9,]       100     0.7755864  2.628621e-04
[10,]       150     0.7778231  3.770554e-04
[11,]       200     0.7775323  3.974573e-04
[12,]       250     0.7672229  2.544826e-04
[13,]       300     0.7613218  2.315345e-04
[14,]       350     0.7561511  1.931951e-04
[15,]       400     0.7550569  2.826986e-04
[16,]       450     0.7490527  3.122957e-04
[17,]       500     0.7445425  2.749257e-04

Quote
Price has the tendency to revert to the mean, or in other words to return to its historic price levels or its averages.

So circling around to add some stats to belatedly back up my original quoted statement (above), I can conclude that as I stated in post #2 in 2007, that price does indeed tend to revert to the mean, but qualify this statement that the reversion is highly dependent upon your forward looking window and MA length as to how frequently this occurs, and whether there is any profit potential in these reversions.

The results looking out 2 * MA Length into the future were actually much higher than I expected both in terms of convergence to the mean % and profit potential from taking all times when price converged. So I re-ran the test again using only a 1 * MA Length as the forward looking period, seeking profit potential from quicker reversions. Here are the results below:

Inserted Code
      MA Length Convergence %           Pnl
 [1,]         5     0.7133251  1.549606e-06
 [2,]        10     0.7002151  6.495704e-06
 [3,]        15     0.6884008 -4.165077e-06
 [4,]        20     0.6842711 -1.026865e-05
 [5,]        30     0.6760218 -1.891711e-05
 [6,]        40     0.6716006 -6.388182e-06
 [7,]        50     0.6714574  1.481646e-05
 [8,]        75     0.6830819  1.185644e-04
 [9,]       100     0.6883740  2.281197e-04
[10,]       150     0.6881545  3.264922e-04
[11,]       200     0.6897978  3.720215e-04
[12,]       250     0.6808357  3.151697e-04
[13,]       300     0.6825144  2.744910e-04
[14,]       350     0.6801142  2.764924e-04
[15,]       400     0.6762715  2.609966e-04
[16,]       450     0.6671049  1.564444e-04
[17,]       500     0.6575977  1.064450e-04

Hmmm, 200 period SMA appears to be the triple crown winner on EU M15 for convergence to the mean in terms of profit potential. A 200 period MA on a 15 minute chart represents 50 hours or just over 2 days of data. Time for me to bow out of this thread.
  • #523
  • Jan 14, 2016 7:07am Jan 14, 2016 7:07am
Ok.
Attached Image
  • #524
  • Jan 27, 2016 1:57am Jan 27, 2016 1:57am
Trading EMA crossovers is an improved form of trading just one moving average. But it wouldn't work always. What I will advise is you add a MACD histogram to it, the MACD will introduce the signal line and measure the relationship. This way you get to see that not actually every crossover is a signal to trade.
  • #525
  • Jan 27, 2016 11:57am Jan 27, 2016 11:57am
Quoting FXEZ
{quote} Good points. I just realized that I posted the 2nd reply to this thread in 2007 to this very old thread! And almost as surprising is that my reply way back then was somewhat sensible. In reply to that post and in reply to your post I decided to test whether the MA is actually representative of "value" and whether price tends to return to these historical value levels and at what frequency for a reasonable forward looking window of 5 times the moving average length. Other forward looking windows could also be used but 5*MALength seemed reasonable...
My hat off to a scientific skeptic
No Highs? No Lows? Must be Bose!
  • #526
  • Jan 27, 2016 12:11pm Jan 27, 2016 12:11pm
MA Crossover Signal Rank from the Most Sensitive to the Dullest
x,y being user-defined lookback window. y>x
Cross=Above or Below

MA0 Cross MAx (current price penetrates above or below MAx)

MAx-MAy≈Constant (high price auto-correlation)

MAx Cross MAy


Enlighten me, when a random chosen summary of price cross another random chosen summary of price, why a signal is somehow generated?
No Highs? No Lows? Must be Bose!
  • #527
  • Jan 27, 2016 1:42pm Jan 27, 2016 1:42pm
Quoting Tardigrade
Enlighten me, when a random chosen summary of price cross another random chosen summary of price, why a signal is somehow generated?
One can demonstrate that the signed distance between the two MAs is directly proportional to the trend slope of the time horizon defined by the slow MA. The histogram of the MACD is a trend-meter. Therefore the crossing is a change of sign of the trend.
No greed. No fear. Just maths.
  • #528
  • Jan 28, 2016 3:12am Jan 28, 2016 3:12am
Quoting PipMeUp
{quote} One can demonstrate that the signed distance between the two MAs is directly proportional to the trend slope of the time horizon defined by the slow MA. The histogram of the MACD is a trend-meter. Therefore the crossing is a change of sign of the trend.
Thank you for your elegant explanation on the inner-mechanics of MA slope change.

I made a bad attempt at raising a rhetorical question, emphasizing on the "random" part; it didn't work
No Highs? No Lows? Must be Bose!
  • #529
  • Apr 22, 2018 1:38am Apr 22, 2018 1:38am
I am using EMA8 and EMA18 crossings for signals. (I am working on a robot to do this for me)
- I use historical pricing on Buy setups to calculate a standard deviation for Stop losses. I set Stop loss at 2 Standard deviation
- I use Historical pricing on Buy Setups to calculate the variance from Price vs. the EMA8. This tells me what the norm is for variance above the EMA.
- I can then take profits and/or tighten stop losses on a spike of price above the EMA8 Line.
- The goal is to also have a graduate tightening of the trailing stop. As price moves up, the trailing stop gradually tightens in hopes that I capture more profits at the top of the trade, prior to the common Pull back prior to the EMA Cross down.

On paper, I can backtest stocks that are non Volatile and I can consistently show profits.
IF you use lower time frames for running the strategy, you have less chance of a big moves where price crosses below EMA prior to crossing of EMA's.
  • #530
  • Last Post: Apr 22, 2018 8:22am Apr 22, 2018 8:22am
Understand some of what you say. I agree, using lower time frames allows you the potential to capture/make more profit.
I simply find that if you go too low of a timeframe (3M or 5M), the cycle distance between crossings is insufficient to cover my spread cost etc..
(I am using a micro acct with high spreads).

The reason why I am thinking it is best to use Close price after crossing to determine the close of trade and buying of new trade, is that
I worry that I worry that the crossing will cross multiple times during up/down movement while the bar is open..resulting in false opens/false closes.
Again, I don't know this but this is just my concern while thinking through the Robot design.

I am searching both Currencies and Stocks and honestly, I can't find one where this strategy doesn't work!
I keep back checking stocks and simply checking buy/sells at each crossing, and even in really tough small/tight crossings, I have small losses or small wins.
Anyways, lets see. Interesting stuff.
0 traders viewing now