Moving Averages

[Chris Nova]

Hi,

I am newly stuck on something I am working on. And so i really need your precious help.
I have a dataset of closing prices of a stock(nbmps) on which i want to calculate the moving averages and the upper and lower bands such as the bollinger band indicator.
I found somewhere few hints about what i am trying to do and i write my code as follow:

gen moveave1= (F1.nbmps+nbmps+L1.nbmps)/4

gen upperband= moveave1*1.04

gen lowerband= moveave 1*0.96


Here's the problem:

line nbmps moveave1 uppperband lowerband Data

when I plot the drawing shows me 4 lines (1 of prices and it's ok) and the other 3 are give or take equal, they do not differ in position, they are overlaid.


I read that usually to generate the upper band somebody use 0.04+SMA.
But i don't know, even because I am a newbie user of stata so maybe I am writing something wrong.

All answers are welcome
Thank for your collaboration

Chris

[Nick Cox]

It's hard to see that you have a real problem here. A moving average with weights 1:2:1 (Hanning) may well be close to the original data. It's really not smoothing very much. Similarly, your extra variables are by design just 4% larger and 4% smaller than the smooth, so why be surprised that they are close?

EDIT: You omitted the factor 2 on the middle value. Dividing the sum of 3 values by 4 makes no sense otherwise.

[Alberto Dorantes]

HI Chris. You can use the filter function from the egenmore commmand by Nic Cox.
Install egenmore from ssc (ssc install egenmore). If your dataset has a variable called price and another time variable called period, you can do the following:

tsset period
egen ma=filter( price), lags(0/3) c(0.25 0.25 0.25 0.25)
tsline period ma

Now you can create other 2 variables for the upper and lower bands by adding/substracting any factor you want. If you want to use the standard deviation for the upper/lower bands, you can use the sd function from egen:
egen sd=sd(price)

I hope this help.

Alberto

[Nick Cox]

Alberto: Thanks for the publicity, but the code in #1 looks fine for smoothing (with the important correction of a factor of 2: see my EDIT please!!!).

Chris appears puzzled that three lines are close together but the values of two variables are by design just 4% larger or smaller.

[Chris Nova]

Thank you for the answers.
As you can notice probably i made some mistakes in calculating the bands. Hence, if there's a package that can I use to exclude my own fault it's better. I am going to try what Alberto said about the egenmore command by Nick Cox, and i will share with you my results.
Thank you both again for the help.



P.S: egen ma=filter( price), lags(0/3) c(0.25 0.25 0.25 0.25)
What the bold part stands for?

P.p.s: Nick you were right i divided by 4 that was nosense, probably i misclick on the keyboard

[Nick Cox]

The filter() function in egenmore was of some use about 15 years ago before Stata introduced better smoothing commands but it has no advantages for you here. I am not disparaging others' work here, as I wrote it.


The only problem I can detect in your approach is that you omitted a factor 2.

gen moveave1= (F1.nbmps+2 * nbmps+L1.nbmps)/4

is fine as a 1:2:1 weighted smoother.

Whether it's good for your purposes I can't say. I don't know what a "bollinger [?Bollinger] band" is.

But it's not at all the same smoother as Alberto is discussing. I imagine he was just giving you an example, and not at all saying "This is the smoother you should use".

[Chris Nova]

Ok thanks that's the final result obtained with your suggestion, it seems great.

Another little question, if i would like to zoom in a specific portion(e.g. 1 oct 15 to 1 jan 16), where can I find the option?

Sorry for the easy question but i didn't find any similar on the web.

Thanks again

[Nick Cox]

Why look on the web? Help is on your system bundled with Stata. For example, the help for tsline gives an example of selecting part of a series from which it follows that something like

tsline nbmps moveave1 upperband lowerband if tin(1oct2015,1jan2016)

is what you want, but you must tsset first if you have not yet done so.

help tsset

[Chris Nova]

Ok, now i can continue my work.
Thank you for your support.

Best regards

Chris