Up next in 10
How to Backtest Technical Indicators in Python - AVWAP vs. ASMA
860 views
Dec 11, 2024
Do you want to learn how to backtest technical indicators in Python? Do you want to find out which indicators are more effective for your trading strategy? If yes, then this video is for you. 👍 Subscribe for more: https://bit.ly/3lLybeP 🔗 Follow along: https://analyzingalpha.com/avwap In this video, I will show you how to use Pandas, a popular Python library for data manipulation and analysis, to run a backtest of any indicator in your Python environment. You will learn how to import data, apply indicators, and evaluate the results of the backtest. I will also compare two pairs of indicators: AVWAP vs. ASMA and VWAP vs. SMA. AVWAP stands for Anchored Volume Weighted Average Price, and ASMA stands for Adaptive Smoothed Moving Average. VWAP stands for Volume Weighted Average Price, and SMA stands for Simple Moving Average. I will demonstrate that AVWAP is not anything special and that it is similar to ASMA in terms of performance. I will also show that VWAP and SMA have different characteristics and that they can be used for different purposes. By the end of this video, you will have a better understanding of how to backtest technical indicators in Python, and how to choose the best indicators for your trading style. If you find this video helpful, please like, share, and subscribe to my channel. Also, don’t forget to hit the bell icon to get notified of my new videos. Thank you for watching, and happy trading! ⏱️ Video Chapters 0:00 Intro 1:24 Get Imports 3:13 Scrape S&P 500 Data
View Video Transcript
0:00
hey there Leo here and I am smiling and I hope you are too you might notice
0:05
there is a new background behind me that is because I've moved and the house you know the reason why there was no content
0:11
for a while is the house we were moving from ended up getting flooded and it just created a nightmare so anyways long
0:17
story short I'm back to create some new content So speaking of content what are we going to learn today well today
0:23
you're going to learn how to test any indicator and see if it has predictive power we're going to use a simple linear
0:29
regression machine learning to figure out whether or not more specifically
0:35
anchored v-wap and the anchored SMA and smas actually have some predictive power
0:41
I know that anchored v-wap is you know thought of very fondly among a lot of
0:47
Traders and I'm not here to knock it I'm just going to show you my results and you know without taking in too much in
0:54
detail they weren't that great right I'm not going to be using the anchor B web but again um you can check out the rest
1:01
of this video or if you're not interested in the code I have another Channel where my next video where I'm just going to discuss how discuss these
1:10
results and how it relates to trading so without dragging this on any further I hope you enjoy this one let me know how
1:16
it goes I'm just resetting up all the equipment and all of that stuff so I'd love to hear any constructive criticism
1:22
or how I could make these videos better alright thanks so the first thing that we're going to want to do is get our required Imports
1:29
and in this instance I'm going to use my own database that I've populated with my
1:34
own price data I teach you how to do that on this channel if you don't have your own database feel free to get the
1:41
data freely from live Finance which I also show you how to do so I'll link the Y Finance tutorial video in the
1:47
description below that way you can follow along with free data if you don't have your own I will get the wired
1:53
Imports reports it all right Imports
2:00
now I'm going to import sys because I need to add my local path so I can
2:05
access my data import sys path append
2:11
home yo Smiggle development Alpha I
2:18
got so this will allow me to access you
2:23
know some of my data that's essentially what I'm doing it I'll point it out exactly when I'm doing it too so we're going to import the standard we'll
2:29
import pandas numpy that's models for the linear regression and P values and
2:35
Pi plot for plotting so import and as PD important numpy SNP import stats models
2:44
as SM import matplotlib dot I plot as PLT and then
2:51
we'll do a start date get a lot of it so we'll do two thousand three zero one zero one an end date
2:57
equals two thousand twenty two twelve Thirty One
3:04
perfect so that will give us our initial Imports now the next thing we need to do
3:11
is get our s p list now this is going to get us the S P 500 list
3:18
quickly and easily create the S P 500 list now keep in mind the reason why we're
3:25
doing this is that we want a quick answer right we're not looking for Perfection here because there are some
3:31
issues with just scraping the S P 500 this like this with um you know various
3:36
biases with look ahead bias and survivorship bias and all that other stuff but for right now we don't really
3:42
care about that we're just trying to identify does this even make sense so we'll just scrape the S P 500 list and
3:48
it should give us you know the answer that we're looking for and if we want to pursue it further we can have a more
3:54
vigorous back test so anyways we'll import requests so that way we can make a
4:01
request to Wikipedia and then we'll use beautiful soup to
4:06
go ahead and manipulate that request so beautiful
4:11
oop and we'll create a function to scrape the S P 500 scrape
4:17
SP 500 list and now what we want to do is pass in the URL
4:25
I have this already punched in and now what we
4:30
want to do is we want to take the response from request.geturl so we make that
4:37
request using this request module and then we pass in the URL and
4:43
we get a response pretty simple right now let me use two parsip the soup equals
4:49
beautiful dupe fonts text we'll do HTML parser
4:58
and now what we want to do is we want to grab the table so table equals soup find table
5:06
and pass it to class so that way you know it knows exactly what we're trying to Target so we
5:13
e-table portable
5:21
and the rows equal table find all ER for table row
5:27
now symbols now now that we have the data
5:33
the rows now we just create a list of the symbols an empty list for row and rows right
5:41
we'll skip the first row because that's a header row symbol equal row find all
5:47
the TD or a table definition first one text strip so symbols append
5:54
symbol then turn tools
6:02
s e 500 list equals sorted support them rape SP 500 list
6:10
and then print SP 500 list and we'll just print the first five
6:16
hopefully I didn't make any typos and of course I did all right so where did I mess up couldn't
6:22
a tree Builder with the features you requested HTML parse or stir
6:28
well makes sense b-a-r-s-e-r
6:34
there we go so now we see that we have you know the S P 500 tickers they're
6:39
fantastic now the next step is to get the prices
6:45
getting the price list is pretty easy again you can use my finance or if you've already built
6:51
your own database at daily prices use the function again so from these are
6:59
uh you know this is my data right from SRC which you notice of course for my
7:06
Civic Alpha AI project where I imported it up here if you're wondering where that is you don't have that unless
7:13
you've named it the same but the point is uh either import using Wi-Fi and answer your own stuff here so funnels
7:19
API import get daily prices
7:24
yeah people get daily prices SP 500
7:29
best start date equals part eight
7:35
and eight eight not the DM that should give me the
7:42
prices and this can take some time because again we've got however many tickers from when we say 2003 now that
7:49
we have our data let's go ahead and create our v-wap and Anchor Point map and anchor
7:57
if you're not sure what view app is I have an article both on avwap and view
8:03
app that teaches you how to calculate and all of the details about that I'm not going to cover that here because we've got 10 steps to go through and I
8:10
don't want to spend the time discussing what V web is if you're here you know the whole point is to learn how to code
8:16
this stuff but if you need the additional assistance just check out the website
8:21
all right so now first we're going to create our two periods and then we need
8:28
a anchor threshold what I mean by that is we're not going to decide hey this
8:33
news should be anchored or this news shouldn't all we're going to do is look for a volume Spike that is three times
8:38
larger than the 20-day moving average and we'll anchor to that point the v-wap
8:44
period equals 20. anchor period equals 20 anchor volume
8:52
multiplier equals three so three x anchor
9:00
period average volume now what we want to do is we want to
9:05
calculate the typical price some v-wap calculations only use close price but
9:10
the proper pendantic or maybe academic way to do it or where the most
9:16
authoritative sites say the typical price which is simply an average of the
9:21
low high and close let's calculate that now the f
9:30
iftical price equals
9:35
f flow plus DF
9:40
I plus f ose and we just divide by three and now
9:48
we need to calculate the volume for that price period so and weighted so EF
9:56
typical price times volume equals the f
10:02
typical price times DF volume right so this price will
10:08
now be weighted by that volume so assuming that all worked which it did now we have all of the requisite stuff
10:15
to be able to create our e-wap calculation so v-wap equals now we're going to group by
10:24
root by but group I thicker now why are we doing that well
10:30
what we don't want to accidentally do is have Apple's volume on uh some other
10:36
ticker's price right so we want to make sure that v-wap is only calculated through each one of these bounds because
10:43
we're you know rolling uh doing like rolling average and we don't want to actually roll our average to the wrong
10:49
ticker so we'll do DF dot root by
10:54
right makes sense because we're grouping by the ticker then we'll take this epx
11:00
volume
11:06
Rolling In The View app period so what are we doing there we're
11:11
grouping by the ticker and taking the volume which will be
11:17
volume TPX volume and then we're adding it now the last 20 periods for rows and
11:25
summing it okay so that's the first part then what we want to do is we want to
11:30
divide that by that price right
11:56
now what we want to do is now that we have that typical price times volume we want to divide by that total volume to
12:04
get that weighted price we'll do DF group I picker
12:12
volume rolling same period
12:17
a period sum that up now what we'll do is we'll
12:23
reset the index level zero and drop it through
12:29
uh so that's just so we can add this and let's go ahead and
12:35
Alpha of course like something up oh fix
12:40
data frame object has no attribute Group by look at this getting out the initial
12:46
YouTube jitters there we go so now we can see that we have this v-wap and this makes sense to
12:52
have these nands here because right we need those 20 values to be able to
12:58
calculate that because it's a rolling 20 period I'll go ahead and delete this and
13:03
now that we have our V web calculated what we'll do is we'll calculate the simple moving average so we can compare
13:09
them right so DF SMA equals the F Group by
13:16
quicker again since we're using that rolling we
13:21
want to make sure that we don't roll across ticker boundaries the rolling d-wap period
13:28
mean right so all we're doing for a simple moving average is we take uh you know the sum of the prices and then we
13:35
just divide by uh you know however many rows there are or take the mean
13:41
and I'm going to reset the index again level zero
13:46
drop if
13:53
now I have okay and now I do the perfect so that calculated for both the SMA now
14:00
we've got a few more things to do we want to calculate the fields where there's that anchor Spike
14:06
and then we want to check them so do that now so d uh anchor volume Spike
14:13
equals the up Group by thicker
14:19
volume
14:24
and we'll do rolling dwap period
14:30
actually I shouldn't do anchor period even in the same anchor period 18 times anchor
14:39
volume multiplier
14:46
that index level zero Property Group
14:51
that's just so we can set right that get that value all right so let's see if
14:57
that's the case so we do is we Group by the Ticker on volume we roll that over
15:04
20 and take the average so we got the average 20-day volume and times the
15:11
multiplier so what this will do is it'll take um you know this is the anchor volume
15:18
requirement right so three times the average rolling volume now all we need
15:23
to do is create a true or false field for if there is an anchor like so anchor
15:34
BF volume is greater than anchor volume
15:41
like make sense then run that no problems and then let's test it what
15:47
we'll do is we'll go to dfdf anchor
15:55
equals through and then what I'll do is I'll just all
16:01
that sweet of them so then we can check to see on these dates whether or not there was indeed a volume Spike now that
16:10
we have the data we can analyze the v-wap versus SMA returns or simple moving average returns and then we'll do
16:16
the anchored B web right because it's a little bit uh A Step Above from in the complexity level so analyze the whap
16:24
versus SMA returns and sorry about this sound I did notice it was chopping
16:29
hopefully this will be a little bit better but anyways what we're going to do here I'm just going to copy and paste
16:35
some code and explain it because I think that would be easier so what we're going to do first because
16:41
we have this multi-index is we're going to reset the index right so that means this multi-index sticker and date they
16:48
will now be columns this then allows us to sort the data frame which it
16:55
shouldn't really unsort it but just to be safe and I put that in there and then
17:01
once we have it sorted we're going to calculate the the daily return and
17:07
that's pretty easy to do all we do is we Group by again because we want to make sure that the percentage change
17:13
is limited by you know that ticker bound
17:19
so we're not calculating like uh Apple's price change with Teslas or whatever right so we create this daily return by
17:26
doing the percent change grouping by the ticker and then how we're going to analyze
17:32
whether or not these indicators meaning the simple moving average or view app has any predictive power is we're going
17:40
to calculate the distance away from price right so we can see here we have
17:46
the simple moving average let's just say it's 101 and the close is you know 100
17:53
divided by 100 so then we can see you know it's essentially one percent away
17:58
right so the idea is the stronger or further away that the SMA or the V web
18:06
is from price it should potentially change that signal that's why I did it that way and then what we'll do the same
18:13
thing with v-wap we take the V web price minus close and divide by close just gets the percentage away from the close
18:20
price positive or negative and then we calculate the return right
18:27
so we can just cumulative sum now you might take issue with this but it's a lot easier to sum some of these things
18:34
and just look for predictive power than to have exponential returns and analyze
18:39
um so again as I said before we're prototyping here we want simple clean and easy we don't necessarily want
18:46
Precision right because we want to say yes this meets our gate criteria we're
18:52
going to dig further into this and then you start digging further into that we'll drop the nand values because there
18:58
will be some nand values because we're doing the percent change right so the
19:04
first one there is no percent change before it right so it'd been in and then we set the index because again we reset
19:11
that index earlier because there's a multi-index back in place and print it out so I'll run this
19:18
and then we can see here we have our multi index data frame once again ohlcv
19:24
our typical price remember that's just the average of the low high and close our typical price times volume so that's
19:32
our volume weighted price and then we get the view app by calculating the TPX
19:38
versus uh the volume for that rolling period we also have the SMA and the
19:45
anchor um so I can't move over there for whatever reason
19:51
uh whatever I'm recording so anyways um so that gives us all of that information
20:05
and then we'll plot the returns let me go ahead and create a new section I
20:11
don't remember what oh fix uh plot returns
20:17
go and now what we're going to do is we're going to get a list of the tickers and
20:23
simply plot all of the returns but we're going to truncate the legend because we don't want a
20:29
bajillion tickers just scrolling down the side of our page right so we already have numpy so we'll say tickers equal DF
20:36
index get level values we're going to want to get the first index which is the ticker
20:42
right and then that would give us a bunch of tickers right so what we want to do is we want to get the unique values and
20:49
then turn it into a list and the reason why we get a bunch is because for every row we get ZTS ETS so we don't we don't
20:55
want that we just want the unique values so now that we have all the tickers let's create a figure and an access so
21:03
uh create figure and access I'm going to do a
21:08
better job with comments so that way if you're looking at this in the future and trying to work through it you'll have
21:13
some more information but anyways big x equals that's the ACT figure which is what the plot is on and then the axes
21:22
um we'll do plot subplot big size and six
21:27
we're not going to have any subplots we're just going to have a bunch of larger plots
21:35
then we'll Loop through each kicker and plot a
21:42
log daily return and of sum
21:47
maybe we'll do yeah um probably doesn't even really matter but anyways we made your log but
21:54
or ticker and tickers thicker data equals
22:00
GF clock picker
22:05
daily return limit of sum right so what we're doing here is we're
22:11
just taking these daily returns for each ticker right so
22:17
um and we're turning it into this the data and then what we do is we want to plot
22:25
for each ticker this data so plot picker
22:31
Theta index particular data label quicker and this
22:38
should give us Right add labels Legend
22:44
this should give us one plot with a bunch of tickers plotted on there and hopefully we don't have
22:50
many outliers but we probably will and that's why I did it with sum versus
22:55
exponential returns but we'll probably still run into the problem but anyways ax set X label
23:02
I want to say this is the date ax that y label
23:07
it says cumulative daily return
23:15
right so this should actually be daily return I'm not going to do the log I'm going to get simple instead
23:21
um and then handles labels equal ax get
23:26
Legend handles labels and then ax Legend handles
23:37
okay now let's analyze results I'll paste this in here and go line by line uh one thing that I think I messed up
23:44
let's scroll back up here uh yeah we'll do stats models
23:51
dot API as SM so I need to fix that
23:56
okay so now what we do is I first create a blank data frame right this is going
24:02
to hold the alpha of the V web and SMA signals Alpha's Ascent also The
24:08
Intercept but basically it just means the return above and beyond what would be expected if you're curious about what
24:16
that means you can check out my website analyzing Alpha to find out exactly what Alpha really is we've got a p-value and
24:23
for both B web and SMA to determine whether or not the results are statistically significant
24:29
now let's do some dropping we drop any nand fields from the data frame and then
24:35
also any ticker that has less than 50 data points I just remove it so that's
24:40
pretty easy right so then the next thing that we're going to do is I'm going to go ahead and loop through every ticker
24:47
right so for every ticker in tickers we get the data remember we have a multi
24:55
you know multi-index so instead of getting this whole thing we'll just get the data for a and dropping the a right
25:01
so just give us gives us the data so then what we do is we then Define our
25:07
independent and dependent variables right after the data cleaning so this is
25:13
that model preparation so what we do is we have the dependent variable or the thing that we're predicting will be the
25:19
daily return technically speaking it's the next day daily return right that's why we see one we start at one here we
25:27
wouldn't want to start at zero because that doesn't really make any sense we want to predict take the V whap and the
25:33
SMA use that data and again to predict the next day and that's why I remove the
25:40
last element from here and start you know essentially remove the first element here so they're equal length but
25:47
essentially what we're doing is we're just saying all right let's take v-wap and SMA distances and see if we can
25:52
predict the next return to some degree and then we fit the model right this
25:58
just does all the calculations for us we create the V web model the SMA model we get the results after fitting the them
26:05
and then we simply plot it and plot the linear regression line and all of that
26:11
other fun stuff so if you're not sure how to do this linear regression thing
26:16
all right I've got a I've got a pretty in-depth tutorial that I thought was really good it teaches you all about the
26:22
different ways to do linear regression and take you through but pretty cool project didn't get a lot of views I
26:29
think it's really good if you're interested check that out that'll teach you all about this and then after the titles what we do is I create
26:36
for every row we append the data right of the alphas that you know the stuff up
26:43
here right so we create a new data frame with called Alpha DF with all that information and I'll run it and you'll
26:50
see here that it starts printing out all of these graphs here's the v-wap equation and the SMA equation
26:57
essentially we're just looking for y you know which is The Intercept or the alpha
27:03
and then this will gonna take a little bit but really what we're interested in here after all of this plotting is that
27:09
we want to understand what is the average or mean Alpha for all of this
27:17
right so we have all of these tickers and now what we want to see is okay does this really give us some results so what
27:25
I'll do now is I'll create a new row here and then run this
27:30
and what this is doing is we're taking essentially this mean row right and
27:36
creating this so we take all of the tickers we just call this mean and then the V web Alpha we take the mean of that
27:44
the SMA Alpha and the reason we're not doing sum or anything like that is that we we want to kind of get rid of some of
27:50
the outliers I mean we could have done median here means find I mean basically again remember we're trying to get good
27:58
right good is good enough it doesn't have to be perfect so let's see that'll take a minute to
28:04
run
28:15
all right so now we have the returns and we can see that here's the tickers now
28:21
there's 500 of them and then we have the v-wap alpha SMA
28:27
Alpha P values and SMA values we see here that the v-wap has a little bit
28:32
higher Alpha than the SMA but we also see that the P values are above 0.5
28:39
which means that this is not statistically significant uh so take
28:45
that with what you will all right so now what we'll do actually before I do that I'm going to create another heading
28:52
analyze a v-wap versus Asma results
28:58
and really what we're testing here is whether or not volume weighting makes sense because AV web is essentially just
29:06
a moving average that is anchored and then I'll paste this in here and we'll
29:12
cover it line by line I felt like that was more effective so the first thing we do is we get all of the tickers that have a volume Spike
29:19
okay and then what we do is we make sure that the ticker has multiple anchors because
29:26
of just how I use the logic down here and there's other ways to do this but again remember we're just trying to get
29:33
quick results so that way we can prototype this put this on our list of things to test further
29:38
we just create some containers for the results and then for every ticker and
29:44
the anchors index we then get the data we use the anchor indices which I
29:51
thought is pretty slick here what we do is for start and IN Zip anchoring to see so what we're doing is we're taking this
29:57
data and then we're looking at between these anchor spikes right or volume
30:04
spikes or 50 days if it ends within 50 days right so basically if the data ends
30:12
before 50 periods or there's another volume Spike you know that that ends but anyways
30:18
um so then we just create the AV web and the Asma using this new data we've seen that before because we have to do it
30:24
between the anchor volume spikes and then essentially you know just
30:30
concatenating all the results together and then here we have this you know
30:35
renaming it and finally merging all the data so when I do this this will give us a data frame of uh you know that
30:43
anchored v-wap and SMA data so now what we want to do is let's just take a quick
30:49
peek and see what does this actually look like on our chart
30:55
right did we actually do the the V web correctly or the anchoring correctly so
31:00
we just create a plot right and then we Loop through you know each one of the tickers for the index
31:07
essentially and what we should see is a graph with the actual chart of a stock
31:14
and then highlighted where the anchoring points occur
31:21
and since there's a lot of data this will take some time so I will see you as soon as it's done processing
31:27
and now we can see here that after this plot we have the price for American Airlines in blue and then we see the I
31:36
guess the volume Spike anchors you know V web and SMA charted here in Orange so it does
31:44
appear you know that looks like that might be a in an earnings announcement well they're
31:49
probably mostly if not all earning earnings announcements but it does appear that it works and the maximum is
31:55
50 days unless another one occurs so looks good to me from a visual inspection so now let's go ahead and see
32:02
if the anchored variants have any Alpha so the first thing we need to do is
32:07
create the distances right we've already done this before so I'll just copy and
32:12
paste this over I feel like this is an easier method at least for me let me know if you like
32:18
like it being more interactive and watching on my typos or copying and pasting me just explaining stuff so
32:24
anyways we create this AV web distance percentage which is just how far away the AV rep is from close and a
32:31
percentage perspective positive or negative same thing with Asma and then we go ahead and drop any Nas and then we
32:38
have our data frame with the AV web and anchored SMA distances right so now what
32:46
we need to do is we need to go ahead and get tickers with AV web right so in the
32:54
next section we will we'll need all of these tickers because we need to Loop through them just like we did you know
33:00
above in this this code way up here right so go ahead and run this
33:07
you can see here are our tickers and now I'll create a new row here and we're
33:14
going to go ahead and paste this in we can see we likely don't need this we already
33:21
have numpy we already have matplotlib because we've already plotted this stuff
33:26
so now what I'm going to do is just run this well actually before I run it I'll explain it right we did touch on this
33:33
pretty in depth but this one the second one obviously we have two new columns we have the SMA Alpha and the smap value
33:41
and we're doing the exact same thing you know we're trying to predict this independent I'm sorry the dependent
33:47
variable and then from a bunch of independent variables right in this case
33:53
we have AV rep Asma and I've added just the simple moving average to see you
33:58
know the anchoring matter at all or should we just be using a 20-day SMA so
34:03
then we fit the models plot it you know we've seen all this before draw straight lines
34:08
uh you know the best fit line then we create this mean row where we take all of the AV web Alpha Asma Alpha SMF and
34:16
and the P values and we just take the mean of that and then run this and you
34:22
can see it'll start graphing quite a few a few things and then this will take a
34:27
minute or two and then once that is done we will go ahead and determine does the
34:33
anchoring actually help us generate Alpha which would be nice for you know
34:38
price earnings gaps or EPS things like those types of setups so with that done processing let's go ahead and analyze
34:46
the results intersection here
34:52
and for context purposes let's get the prior results of those Alpha DF mean if
34:57
we recall they were very similar but they did not have significance because
35:02
you know they're above that 0.05 threshold so let's go ahead and see if the anchored variants are any better so
35:09
Alpha DF 2 mean and here's the big reveal so it looks
35:15
like the AV wrap performs a little bit less than the anchored SMA value and
35:23
then the simple moving average performs the best right so it seems like the
35:29
volume weighting or the anchoring actually hurts the indicators potential profitability and and then we have you
35:38
know no significance here so what does this mean right so when you're looking at data you have to figure out okay why
35:43
is it the way it is right and try to hypothesize that well my thought process in regards to anchoring is I don't on
35:50
volume is I don't think it's that great of an idea because if you really think about Price Right Price is a discounting
35:57
mechanism where you're essentially you know try you know the market is very
36:02
good at getting the latest information so a lot of times when you anchor on a
36:07
lower price let's just say in a post earnings announcement drift scenario your actually using and heavily waiting
36:15
volume that occurred earlier on in the period where price hasn't gone through
36:21
that price Discovery yet so my initial thought process you know when thinking
36:26
through this before doing this was that you know what anchoring is probably not that great of an idea because you're
36:31
overweighting historical prices relative to newer prices which are in my opinion
36:39
and the Market's opinion you can't argue with it right I mean it's incorporated as you know that that further
36:44
information so I think this what makes logical sense I'm not going to use anchoring and my strategies how about
36:52
you what did you think did you like this video and if you did let me know between if you like me typing stuff out or if
36:59
you like me you know just explaining the code and hopefully I'll see you in the next video thanks bye
#Investing
#Programming