Holt Winter's Double Exponential Smoothing.

[r4bb1t 10]

Hello Traders.

i've seen that someone implemented Holt Winter's double exponential smoothing indicator. which is used in business forecasting.

http://www.forex-tsd.com/blogs/newdigital/494-new-elite-section-indcators-holt-double-exponential-smoothing-indicators.html

and here's simple reference of it. http://vbautomation.110mb.com/FORECASTING/expoential_smoothing_methods.htm#Holt's

can anyone convert it to Easy Language please?

[Tams 20]

can you describe what it is and how to use it?

[r4bb1t 10]

can you describe what it is and how to use it?

 

Hi Tams.

it's kind of exponential smoothing. since normal EMA has an alpha factor that is determined as 2/(Period+1).

Holt Winter's method uses 2 factors called alpha and beta. the important factor is beta which weights more when data has trend. though it's not adaptive.

therefore we can expect that Holt Winter's method would fit better than normal EMA when market moves in a trend.

it's all i know. for mathematical reference please check here. http://www.scss.tcd.ie/Rozenn.Dahyot/ST3007/4DES.pdf

[Tradewinds 12]

This code does not work. It's not tested. It's not meant to be complete. It's totally wrong. It may be a pathetic and foolish attempt to help. So why am I posting it? Because I just like to make an effort and start somewhere. Maybe it will be a starting point. I don't want to do all the work myself, but I thought I'd make a feeble attempt just for fun. Sometimes getting started is the most difficult part.

 

input: Alpha(0.2), BetaHolt(0.3);
var: Level(0), Slope(0), SlopeInitial(0);
var: Y(0), F(0), k(0); // Y = price
var: InitialLevel(0);
var: Iterations(9);

Y=Close;
SlopeInitial = Y - Y[1];

InitialLevel=Y;

// Loop through for number of Interations

Level = (Alpha * Y) + ((1 - Alpha) *(Level[1]+BetaHolt[1]));
Slope = (BetaHolt * (Level-Level[1])) + ((1-BetaHolt)*Slope[1]);
F= Level + Slope;

F = Level + (k * Slope);

plot1(F, "DES");

[r4bb1t 10]

Because I just like to make an effort and start somewhere. Maybe it will be a starting point

 

Hi Tradewinds.

Thanks for paving the way. i'm also struggling by myself.

there's an excel worksheet file in the link of my first post.

seems it requires different calculation for initial iteration.

btw in your code. you're plotting Forecast line. not the smoothing line. (Level)

and i've noticed that when i replace alpha to 2/(Period+1) and beta to 0. it's exactly same with Normal EMA.

so the rest part to accomplishing it is beta, anyone join us?

Edited September 11, 2011 by r4bb1t

[r4bb1t 10]

by the way i've coded Brown's Double exponential smoothing in the reference document.

it's not the DEMA. it's slower than normal EMA and smoother. it's just enclosed EMA with EMA.

i've plotted normal EMA(Red) and BDES(Blue) together.

now, let's complete Holt Winter's too

 

 

#Brown's Double Exponential Smoothing
#Function Name : BDES

Inputs : Price(NumericSeries), Period(NumericSimple);
Vars : SES(0), Alpha(0);

Alpha = 2/(Period+1);	#same as normal EMA

IF Index==0 Then {
   SES=C;		#Single Exponential Smoothing
   BDES=C;		#Brown's Double Exponential Smoothing
} Else {
   SES=(Alpha*C)+((1-Alpha)*SES[1]);
   BDES=(Alpha*SES)+((1-Alpha)*BDES[1]);
}

Edited September 11, 2011 by r4bb1t

[Tradewinds 12]

btw in your code. you're plotting Forecast line. not the smoothing line. (Level)

 

Oh! :rofl: See, I learned something new. If I hadn't shared my imperfect code, I wouldn't have learned that. So what is the Forecast variable for anyway? What role does it play?

[r4bb1t 10]

So what is the Forecast variable for anyway? What role does it play?

 

my understanding is that the Forecast variable supposed to be precede data like link below.

but i've noticed that DEMA or ZLEMA came from brown's double exponential smoothing. it's the brown's Forecast variable.

i guess pic in the tsd link might be Holt Winter's forecast variable that made to accompany with price like ZLEMA does

 

https://www.student.gsu.edu/~lcross5/Forecasting%20Assignment.htm