The Finance Book Club

Chapter 2 - SMAs, EMAs or Other?
Ok, let’s just focus on the basics.

1. Simple moving averages (SMA) are nonrecursive filters, where all the coefficients equal one divided by the number of elements in the filter
2. When time advances by one bar, the fractional part of the oldest data is discarded and the newest data is added; this is the ‘moving’ part
3. "As simple as it gets!"
4. It’s a bit like a ‘window’ that passes across a data stream
5. An impulse
   1. A theoretical data object
   2. Zero width and infinite height
   3. Used to assess the filter transfer response
   4. Think of it as the clapper in bell
   5. The bell is the transfer response; the ringing is the output
   6. The ringing is the impulse response

6. You can approximate an impulse as a data stream with a value only at one point, and zeros in all other data samples
7. Sliding an SMA window across the impulse data stream, filter will only have an output when the impulse falls within the filter window
8. This makes it a special case of the FIR (finite impulse response) filter.(ok, but what does this have to do with anything??)
9. The SMA is synonymous with finding a best-fitting straight line to the data, when lag is disregarded. Ehlers finds this ‘fascinating’.
10. SMA lag is (N-1)/2
11. Critical period
    1. Of a filter output is the frequency at which the output power of the filter is ½ the power of the input wave
    2. The critical period of an SMA is approximately 2x its length; which defines its ‘pass band’

EMAs vs. WMAs

1. EMA should always be written as EMA = alpha * Input + (1- alpha) * EMA[1] to avoid computational errors; where alpha = the value of the first bar
2. Ehlers is fond of saying things like “It is instructive to examine the EMA response to a step function.” But is it really? Instructive of what?
3. Ehlers disparages the WMA (weighted moving average) as the invention of a trader who didn’t understand that the WMA has poor rejection in its transfer response; which is to say it’s laggy and offers no benefits over an EMA

Median filters

1. A median filter simply picks the middle value of a sample set of data after the samples have.

Median filters are best used when the data contains impulsive noise or when there are wild variations. Smoothing volume data is a good application for a median filter.

Should I now test an EMA vs a WMA? I'm inclined not to as there is much testing to be done in later chapters. Also I don't really care about the minute advantages of one moving average over another. They are, as Ehlers would admit, all just filters of varying characters and predictive of nothing.

Hi clemmo, imo your inclination is probably sound

Here's a chart I did up on the fly one day, where I placed a group of 21-period moving averages (random choice)
plus a 21-period donchian channel to see what it looked like on a chart period I was familiar with.

None of the five same period indis probably saved a trader caught in a 50 pip whipsaw fake'n'flip later on the top right hand part of the chart lol.
(Plus I include the exploded chart beneath it of that top right hand part Fake'N'Flip also to show where it happened during that day)

Indis are a great tool imo.
But like everything offered in Forex thay have their own limitations also. Once we understand their limitations, we can utilise their 'adavantages'.

all the best



(PS if you come across a Book called "Time Compression Trading - Jason Janovsky", I would be interested in your summary sir,
I sped-read it lol years ago but didn't get the fuller picture, I also don't use certain forex phrases he uses, but I did get something from it in my small view)

.

I'm really glad to see someone else's analysis here. I still would even if we didn't agree. I will add Janovsky's book to the stack. Thanks!

TA is not a viable way to get an edge for trading. Page 37.

Consensus changes are traded by nett winning traders (weekly, daily) than nett losing traders (5 min 15 min etc). That's what I got from it. For forex his open interest and volume being the 3rd and 2nd most important things to look at - useless. Perhaps there is a proxy to use or forex futures data. His previous book leads to this one where he gets critical of TA.

Cheers..

Hi FWR, yes I was hoping for a fuller summary from clemmo but you have quoted correctly from his 200+ page book.
Ironically though, we can still observe TA at the points where he suggests "compression" is occuring. But it is NOT the edge for trading I would agree.
That imho still lies within us, the trader.
We can discuss this further should the book make the 'play list' here
all the best

Yes of course TA helps. It's all about what one is trying to achieve. Although the book was worth reading, just like all US stuff so options and futures centric though.
All the best.

Hi FWR,
Well whether TA helps has been the subject of many a thread on FF here
But it is definately observable imho.
I had difficulty with the book but I did smile at the part where he suggests WE do not determine our own winning trades, other traders do.
But I might leave it there
Cheers pal

Chapter 3 - Smoothing Filters on Steroids

1. The objective of smoothing filters is to get the most smoothing with the least lag
2. None of these filters are predictive
3. For a nonrecursive filter having N elements, the lag of that filter will be (N − 1) / 2.
4. SMAs are a least-squares best fit to a straight line
5. Nonrecursive filters can also have coefficients that are a least-squares best fit to a quartic. A quartic is a polynomial of 4th degree (power).
6. The best filter for most smoothing applications is a two-pole SuperSmoother.
7. Ehlers likes the superSmoother and recommends it for removing ‘aliasing’ noise. Which I assume is noise. Like, random market fluctuations.

Once again, I could get you a superSmoother indicator, and compare it to some less exotic moving averages, but I think I will skip it for now. Ehlers is leading up to something special and like any good engineer he has structured the book chapters to methodically show how his masterpiece is constructed from conception to reality.

I do seem to recall that the SuperSmoothers I've tested out in real trading did not do substantially more for me than say, the Hull Moving Average which was a favorite for a while.

Chapter 4 - Decyclers

1. The corollary of the detrender is the decycler. It removes the cycle components and retains only the trend.
2. Exceptionally low lag
3. Ends up simply producing something like a trend line - an ‘instantaneous trendline’
4. Decyclers have less lag than superSmoothers but are worse at removing aliasing noise
5. Decycler Oscillator - a trendline that displays as an oscillator
6. Decyclers function the same as a low-pass filter
7. They are created by subtracting the output of a high-pass filter from the input, removing the high frequency components by cancellation

Ok, these are definitely worth seeing in action. I will go on a side quest to test these out, time permitting tomorrow.

Decycler Examination
Let's start with the Decycler.

Here's what Ehlers has to say about it.
"When the results of the decycler are compared to those of the SuperSmoother, the decycler will always have less lag. However, a decycler is only a one-pole filter and therefore has inferior filtering capabilities. Therefore, a decycler should not be used as a smoothing filter to remove aliasing noise. Rather, its role should be relegated to producing an instantaneous trend line where the selected cutoff period is relatively large. The large cutoff period enables the decycler to attenuate the aliasing noise because it is many octaves away from the Nyquist frequency."

I have attached a Supersmoother and a Decycler so that you can follow along at home if you like. I believe most of these indicators were made by Mladen or his followers over at Forex-TSD back in the day (2000-2010??). Ehlers offers up the code for everything in his book and while there is some room for creativity/error in interpretation when porting over to MQL, I think we can be fairly confident we're looking at the same thing.

I turned off the bands in the decycler and turned off the line signals.

Here are the settings for the decycler.

For the SuperSmoother

True to what he says it does seem like the smoother is better at smoothing, the decycler is more reactive, even though it has a filter length twice the value of the decycler. Blue line is the decycler. Green is the smoother.
Ok, so now what?

FYI

I've turned on the line signals, overlaid the Decycler (blue/yellow) on a chart with a high-pass filter length of 30 which is the setting Ehlers uses when measuring the cutoff lag. I've also added an EMA (red/yellow-green) with a period of 30.

Ehlers doesn't show us a chart using the Decycler. He's mainly focused on the ultra-low lag. He mentions that in a 20-bar cycle period the lag is ~ 1.5 bars.

So what do we see?

It seems like the Decycler is far more responsive than the EMA. This is both a blessing and a curse. It would get you in and out of a trade much faster, but results in more false signals.

I can turn up the decycler filter to 140 before it starts to get as slow as the EMA.
Does this matter though? Maybe it is slightly ahead of the EMA and generates slightly fewer false signals?

So I think we can confirm that this would be a much faster way to detect reversals than an EMA. This is as bold a claim as Ehlers ever makes about the Decycler. This is the description from a 2015 issue of Tradestation's 'Trader's Tips'.
Now I changed the EMA to an SMA. It's even slower but far fewer fakeouts.

On H1 you can see that this would be nearly un-tradeable, even with a very large filter period.
I suppose it really depends on how you like to use a moving average. What else should I try?

This is 20 periods on H1 for both. I think it's too choppy using either signal, but the EMA at least keeps us in some longer moves.

One of the features on many Mladen-esque indicators is a filter for the price values themselves. I'm not sure if this is an Ehlers invention; it seems like it would be. I haven't yet come across it in this book. When I activate that and set it to 1.0 I get something that almost looks workable. But don't forget that there is a 1-bar lag, minimum, meaning for historical backtesting we have to consider the bar that closes after the signal change to be the price location. When you work this out it ends up being slightly less impressive.

If you want to trade intra-bar signals, I think that way leads to madness, but be my guest.

I guess, let's look at the Decycler Oscillator now?

Cheers Parisboy, thank you, sir

Decycler Oscillator

Ehlers says,
"The decycler oscillator can be useful for determining the transition between uptrends and downtrends by the crossing of the zero line. Alternatively, the changes of slope of the decycler oscillator are easier to identify than the changes in slope of the original decycler. Optimum cutoff periods can easily be found by experimentation."
Yes, lots and lots of experimentation.

I lied when I said there was no decycler chart, here it is on the main chart window. The sub-chart window shows the oscillator. Ehlers uses 60 periods for the oscillator and 30 for the decycler on a D1 chart of Dollar General. It seems to me if you tried to trade the downward zero-crosses in this uptrending instrument you would regret it. Instead, they might be useful for finding re-entries for a bullish bias. Then, later in the year when the trend declines, you would want to reverse this policy; though there isn't enough data to say for sure.

Here is my reconstruction of Ehlers' setup. 60/30 periods on a daily chart of EURUSD.

Once again we're on our own to try and decide how to make use of this tool. Ehlers does have a general trading advice chapter at the end of the book which we'll definitely review later. For now though, let's skip ahead and read what he says about using oscillators.

"I think the best way to trade an oscillator is to anticipate the turning point and rely on the statistical return to the mean as being predictive. [...] In contrast to the conventional wisdom, the Stochastic indicator anticipation of the turning point trading rule would be “make a long position trade when the oscillator crosses below 20 percent, and reverse to a short position trade when the oscillator crosses above 80 percent.” The anticipation rule enables us to get the trading signal nominally three bars before the turning point of the indicator. We still have the four bars of computational lag and the one bar lag of entry following the signal. Therefore, our net lag in making the trade entry is only two bars relative to the real turning point in the data. In other words, out timing is just about right."

Ehlers calls this the 'anticipation rule'. Unfortunately we don't have an 80/20 turning point to use with the decycler oscillator. Instead, let's mark out the zero crossings and see what would happen if we traded them without any guile.
This looks like too many signals.

As expected, it does quite well in trends and gets chopped up in ranges. Maybe guile > anticipation? I suspect the account balance will be about even not including trading costs. If we had a notion about the overall trend it would have been better to use counter-trend signals as waiting or re-entering periods. It also would be interesting to add a rule about what to do when price crosses the Decycler line on the chart, especially when trading countertrend.

Still, we mustn't judge too harshly at this stage. Ehlers, as I mentioned, is building up to something. "Iterating", I think the whiz kids call it.

let's be clear every body has his beloved Moving Average (Settings). I will not participate to the game "my moving average is better than yours"

The problem is most people do not know what is a MA and how to use it for Analysis and Trading.

Ehlers is not a trader, he mainly writes books, (challenging, stimulating) but of very few interest for trading.

Moreover his pet idea is that to reduce the "lag" of a MA by using sophisticated unapplicable by most people or/ and platforms.

Let'go back to Moving Average as shown by Ehlers and Clemmo17 a Moving Average is a FILTER.

But what does it filter ? How to use this filter to trade ?

Clemmo17 shows us a chart with a 20 period (Time Unit = 1 Hour)

This means concretely for a trader that ;the 20 period moving average

a )reduces to zero the presence and the magnitude of the 20 frequency if it exists in the data series (the Price Action)
b )drastically reduced all shorter frequencies , but they may show some traces of their presence in the 20 period MA
c) all frequencies greater than 20 period are present in the 20 period MA. The longer the period the more completely the full magnitude of the component come through

As you can see :

using a Centered Moving Average 32 Time Units (in blue)

This Moving Average is in the CENTER of Price Action indicating the TREND

Price Action fluctuates / oscillates around the Trend / Moving Average in a Sinewave form

The Areas ABOVE and UNDER the Trend / Moving Average , summed , are more or less equals.

To quote Ehlers and use his concepts Price Action is simultaneously in a Trending Mode and in a Cyclic Mode

I don't recall seeing this before. I wonder if it can go some way to explaining why, in my experience, the profits from a trend-trading system are usually closely balanced by the losses from trading that system in a range. The reverse is also true. The profits from trading a range-trading system will be wiped out nearly to the dollar by a trend before it begins to range once again.

Clemmo17 I do not want to hijack your thread on the Moving Average subject with Jim Hurst methodologies . Hence I think it is better for the readers (me included) that you could end your excellent summarizing of Ehlers book.without being interrupted.