YT YouTube2Text

YouTube2Text → Transcripts → Pairs-trading strategy from scratch in Python

Pairs-trading strategy from scratch in Python — Transcript

by Algo-trading with Saleh · 6,924 words · 911 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00today I'm going to write a Paris trading
0:02strategy which is the basic strategy
0:04that Edward thorb used to make his 800
0:07million Fortune I've been obsessed with
0:09this strategy ever since I read his book
0:11called A man for all markets a few years
0:13ago and realizing that not only this guy
0:16beat the casino by inventing card
0:18counting but he also beat the stock
0:20market with the strategy not only his
0:22fund was profitable almost every single
0:24month but his worst monthly loss was
0:26less than 1% it's a market neutal
0:29strategy which means potentially we can
0:31make money whether if the price is in an
0:33uptrend A downtrend or even in a Range
0:35I'm going to show it on a chart in a
0:36second but then I will also show the
0:38basic math behind it in a Jupiter
0:40notebook and then we will write it
0:42inside Jesse so we can run some back
0:44test and see some actual results so if
0:46that sounds interesting let's get right
0:48into
0:49it all right let's begin with a very
0:51simple example ethereum versus ethereum
0:54classic they have different price ranges
0:56ethereum is in matter of thousands of
0:59dollars but ethereum classic is I think
1:01in hundreds but we know their prices are
1:04highly correlated and that's really what
1:06we care about so for example on this
1:08chart I have both their prizes and I've
1:11set it to be on the same priz scale just
1:13so that we can see them easier and you
1:16can see for example in here the price of
1:18ethereum is actually lower than the
1:20price of ethereum classic but it becomes
1:23the opposite in here the price of
1:25ethereum is actually more than the price
1:27of ethereum classic so what does it mean
1:29it means that if we open the long
1:31position for ethereum at this point in
1:33time and at the same time open the short
1:36position for ethereum classic here and
1:38hold that position when we reach this
1:40point in time you can see that it's the
1:43opposite now the price of ethereum
1:45classic is lower than the price of
1:47ethereum so what does it mean it means
1:48that yes both the prices are lower than
1:51what it was at this point in time but if
1:53we opened two opposite side positions
1:56one long and the other one short at this
1:59point we would have made money why
2:01because our short position would have
2:03made more money than the amount that our
2:06long position lost so the difference
2:08between the panl of both positions would
2:11be our profit now if I go back in time
2:13you can find so many positions where
2:15this opportunity existed I'm going to
2:17explain the math behind this strategy in
2:19a second with some code but I want to
2:21mention that this video right here by
2:23copian is very initially learned about
2:26the math behind it and in my video I
2:28won't go to as much detail as they did
2:31so definitely check out their video if
2:33you want to learn about this topic in
2:35more details all right so first I'm
2:37going to show you the basics of parist
2:39trading in a Jupiter notebooks which is
2:40great for a presentation and in case you
2:42didn't know if you want to use jupyter
2:44notebooks with Jesse all you need to do
2:46is to make sure they are inside your
2:48Jesse project all right so my project is
2:50named but I have multiple Jupiter
2:53notebooks and this one is the one I'm
2:55going to use all right so first we need
2:57to select the kernel and I'm going to to
2:59select one which already has Jessi
3:02installed in it all right let's add a
3:04code block and first I'm going to import
3:06all the necessary packages so I'm going
3:08to say from Jesse import utils and the
3:12research module now the utils will give
3:15us some utility functions and the
3:16research module is used for importing
3:19data from the database or even running
3:20back test but I'm not going to use that
3:22in jupyter notebooks in this video but
3:24you can if you like to all right so next
3:27I'm going to import JC helpers as JH
3:30which will give me some helpers that
3:32aren't really used in most of the
3:34strategies but I'm going to use it in
3:36here next I will import numpy as MP from
3:40daytime I'm going to import daytime and
3:43lastly I'm going to import the Matt plot
3:46Li library because I'm going to show you
3:49some charts all right next I want to
3:50import some data and I'm going to use
3:52the research modules get candles
3:54function and first it will ask me the
3:57name of the exchange all right so let's
3:59def find the exchange variable here I'm
4:02going to use binance Perpetual Futures
4:05and here I will pass the exchange the
4:08second one is the symbol and I want to
4:10use eth and Etc which is ethereum
4:13classic all right so the first one is
4:15going to be e us and the time frame I
4:18want it to be 15 minutes and then we
4:21have the assort now the assort date that
4:24is accepted here has to be in time stamp
4:26format and to convert that I'm going to
4:29use use the helper function from Jesse
4:31so I'm going to say this start date
4:34equals and I'm going to use date to Tim
4:36stamp and in it I will pass the date
4:40since
4:422024 the month could be the 10th month
4:45and since the first day and for my
4:48finished date I'm going to give it again
4:51date to timestamp again 2024 10th month
4:55and the third day so I only want data
4:58for 2 days B basically all right so now
5:01I have the start date right and the
5:04Finish date all right so and I can leave
5:07the other ones all right now this
5:08function will return two values the
5:10warm-up candles and the actual candles
5:12that I want and because I don't intend
5:14to run back test with it in this Jupiter
5:17notebook I don't care about the value
5:19for the V candles right so I don't want
5:21the first one and that's why I'm using
5:23this underline value here but the second
5:25one I will call it C1 or candle one and
5:28there we go now for the second one we're
5:31going to do the same thing right except
5:34instead of eth we want it to be Etc all
5:36right let's run this and it's all good
5:38now next because I want to show you a
5:40chart of these values I'm going to need
5:42time stamps so I'm going to create an
5:45empty list called times and I'm going to
5:48Loop through the first set of candles
5:51right and in it I will simply say times
5:53do pend and date time do from time stamp
5:57and then I will pass the time stamp of
5:59each candle and that would be the first
6:01index but because this function expects
6:03the time samps to be in seconds but
6:05Jess's times samps are in milliseconds
6:08I'm going to have to divide this by
6:111,000 all right there's one more thing I
6:12need to do just like this chart that I
6:14just showed you which both prices were
6:16on the same price scale because they
6:18weren't absolute values of prices they
6:20were price changes in matter of
6:23percentage right so we need to do the
6:25same in our python code so I'm going to
6:27say C1 equals use
6:30do prizes to returns and this is an
6:33already built-in function in Jesse so I
6:35don't have to Define it from scratch
6:37which is really nice so in it I'm going
6:39to have to pass all the candles but I
6:42only want the closing prizes and that
6:44would be C1 and I'm going to select all
6:47the candles but then I'm going to select
6:49only the closing candles and that would
6:51be the index of two so in case you
6:53didn't know the zero index in Jesse is
6:55timestamp and then we have the open
6:58close high low and then volume so the
7:00close would have an index of two and
7:02then I'm going to do the same thing for
7:04the second price data or ethereum
7:06classic right and we should also change
7:08this all right now we can run this one
7:10okay we're getting an error all right
7:12this shouldn't have been there it should
7:14have been inside here all right so now
7:15we can finally plot the actual price
7:17data and for that I'm going to say PLT
7:20figure which is used to set the size of
7:22the chart and in it I'm going to say fix
7:24size equals 15x 6 now this Size Doesn't
7:27Really Matter much it's just for
7:29presentation you can set whatever value
7:31you like next I'm going to say PLT do
7:33plot and I'm going to pass the time
7:36stamps and then the actual price data
7:39and for color I'm going to say blue and
7:42the label is going to be e next I will
7:45do the same thing for ethereum classic
7:47and I'm going to set the color as red
7:49and this should be TC and finally I can
7:52do pl. Legend to show me everything all
7:55right so now we can see the price data
7:57for both the assets on the same chart
8:00which is really cool now if you pay
8:02close attention you can see that our
8:04Theory very much sounds right in here
8:07for example the price of eth was lower
8:09than Etc but in here it was higher than
8:13that so in other words if we open the
8:15long position here and shorted the price
8:18of Etc and did the opposite in here we
8:21would have made money and the same goes
8:23for so many other places it goes here it
8:26goes here here here so it just continues
8:29right which is very promising all right
8:30so now that we have both of them on the
8:33same chart we should begin writing the
8:35actual code based on math that we can
8:37actually use to trade it right because
8:40we're not going to have the chart when
8:42we are trading this with a bot right now
8:44we're looking at this chart with our
8:46eyes that's how we know this happening
8:48so we need to use statistics to come up
8:50with a good strategy the first step is
8:52that we need the spread in other words
8:55we need the difference between the two
8:56price data so I'm going to define a
8:59spread as C1 and then I'm going to
9:02select since the first candle up until
9:04the end and the reason I'm doing this is
9:06how we defined this C1 variable we
9:09defined it in here we converted the
9:11candles into returns and when we did
9:14that the First Data actually became a
9:16nan value and that's why I'm doing this
9:18and if you don't believe me I can print
9:20this data so you can see it so let's
9:22print the first value and you see it is
9:25Nan right and we cannot really have that
9:27in our strategy so that's why I'm going
9:30to select since the first data and when
9:33I say first I mean the index of one
9:35because the actual first value would
9:37have the index of zero so we're going to
9:39select this one until the end minus C2
9:43and again selecting since the first
9:45index up until the end right so this is
9:47going to be our spread which is the
9:49difference between the two price data
9:51now to actually take advantage of this
9:53data having the spread alone isn't
9:55enough we need the zcore of it now what
9:58is zcore the Z score is the price
10:00difference or the spread that we just
10:02defined compared to the mean of the
10:04price data and that's what we really
10:05care about so we're going to define the
10:07zscore now Jesse already has a utility
10:10function which will calculate the zscore
10:12for me so I'm going to use that one it's
10:14called zscore and it takes the series
10:17data or in this case it would be our
10:19spread now I'm going to also Define the
10:21mean now by the way the zcore function
10:24it will use the mean itself if you're
10:26curious to know how it works is it's
10:28basically this price series minus the
10:31mean of that price series divided by its
10:34standard deviation right now you don't
10:36have to know about the exact math behind
10:38it you could just use it but knowing it
10:40certainly helps all right so I am also
10:43going to define the mean of it myself
10:45because I want to print it on the chart
10:47and for that we can use numpy and we
10:49will just pass the zscore and now we can
10:52plot them right so first I'm going to
10:55say PLT
10:57figure and then pass the fix size just
11:00like before 15 and six next I'm going to
11:03say PLT plot and I'm going to pass the
11:07times again selecting since since the
11:09index of one to SK skip that Nan value
11:12and then I'm going to pass the zcore I'm
11:14going to set the color to Blue and the
11:17label could also be Z score next we're
11:19going to plot a horizontal line using
11:21this index and I'm going to pass the
11:23mean that we just defined here I'm going
11:26to use the color black and set the line
11:28Style to Dash and the label is going to
11:32be mean and then I'll do it two more
11:34times for the values of 1.2 and minus
11:381.2 with colors of red and green all
11:41right so let's see how this looks like
11:43and there we go so now we have the zore
11:45of the two price series and we can see
11:48how their correlation or the related
11:52price is actually moving so for example
11:54in here we can long the first asset and
11:56short the other one and when it reaches
11:59the mean we will just close the
12:01positions and we can do the opposite
12:03when the zscore goes above this line we
12:05can short the first asset and long the
12:08second one and again when it reaches the
12:10mean we can just close it right and and
12:13as you can see we had opportunities for
12:15this type of trade in more than once we
12:17had it here we had it here here here
12:21here here and so much more right so it
12:24looks very promising but here's the
12:26thing though right now we have this data
12:29honest beautiful chart which is only
12:31showing us the price data for two days
12:33but in reality we're not going to have
12:35charts we're going to trade this with a
12:37butt right so we need a statistical way
12:40to write all of these things but not
12:42only that we need to make sure these
12:44prizes are actually cointegrated now
12:46what is congration well again this video
12:49will describe it so much better but I'm
12:51also going to try my best you see to
12:53trade any mean verion strategy we need
12:55that price data to be stationary and
12:57what is a stationary well it means that
13:00the mean of that price data or data
13:03series is not moving now what is a
13:04stationary it means that the mean of the
13:07price data or data series is stable it's
13:11not changing because right now in this
13:13picture we know the mean is this but
13:16that's because we already have the price
13:18data for 2 days right so for example at
13:21this point in time the mean wouldn't
13:24exactly have been this value or at this
13:26point in time it wouldn't exactly have
13:28been that because we would would have
13:29calculated based on this much data right
13:31but now that we have all of it we
13:33already know what the mean of it is
13:35right so if I try to calculate the mean
13:38of this data series is going to change
13:40now we don't want it to be like stable
13:43100% to be able to trade but the thing
13:45is if we try to trade a mean verion
13:47strategy on let's say just one asset we
13:50know for sure that it is not going to be
13:52a stationary however we can still trade
13:54them if they are conr and what does
13:57congration mean it means that yes these
14:00two price data even though individually
14:02they're not stationary some kind of
14:04combination of their price data which in
14:07this case would be the zcore that we
14:09just calculated that series is a
14:11stationary so that's what it means right
14:14and if that's the case then we can trade
14:16them against each other but if they're
14:18not we just can't now there are some
14:20statistical ways to calculate this but
14:22again Jesse gives us a utility function
14:24which will take care of this so I'm
14:26going to use that one so let's define a
14:28new variable called is cointegrated and
14:32in it we're going to use the r
14:34cointegrated utility function and first
14:37we will pass the first priz data and
14:38then the second now let's print this and
14:41it is returning true so now we know they
14:44are qur now if you're curious to know
14:46the source code of this well here it is
14:48we're using the set model module of
14:51python and then we use the P value and
14:54if it is below the cut of value we say
14:56it is cointegrated otherwise it's not
14:59now I know this sounds confusing but
15:01again watch that video of copian and
15:04hopefully it should make it clearer
15:06right now let's move on to Jesse so we
15:08can write the code in it so we can run
15:10some backst and see some actual numbers
15:12all right I'm going to start by
15:13generating a new strategy and I'm going
15:15to call it P
15:18trading all right so now I'm going to
15:20copy this and I will open my CER editor
15:24which uses Ai and is so much smarter and
15:26faster to use and I'm going to look for
15:29for this all right so this is the
15:30generated code now the real challenge
15:32for this strategy is that we cannot only
15:35Define it inside one strategy file
15:38because if I try to run a back test you
15:41see we have these routes and right now
15:43it is set to BC so let's change this to
15:46East UCC and the time frame to 15
15:50minutes and this one to P trading all
15:54right and the duration is since
15:56beginning of this year until just if you
15:59days ago now here's the thing though we
16:00need to trade this for both eth and
16:03ethereum classic so I need to add
16:05another trading route and for this one
16:07I'm going to choose Etc and the same
16:11time frame but now we need to also pick
16:13another strategy so picking this one
16:16alone won't work so how do we do this
16:19now I need to generate another strategy
16:21and I can pick whichever name I like but
16:23I'm going to go with Paris trading two
16:28all right so I'm also going to open this
16:31one all right so now we're going to deal
16:34with two strategy files now I know you
16:36guys have some questions like how do we
16:38communicate between the two strategies
16:40how do we connect them to each other and
16:42I'm going to take care of the heavy
16:43lifting right now so let's connect these
16:45two strategies together so inside my
16:48first route I'm going to define the
16:51before function and in it I'm going to
16:53say if the current index is zero which
16:56means it's the beginning of the
16:57execution of the Strat
16:59I'm going to initiate two variables and
17:03I'm using this special dictionary
17:05variable inside Jesse called shared
17:07wordss now why is it called that it's
17:10called shared wordss as in whichever
17:11variable that you define in this
17:13dictionary is going to be shared among
17:16all the trading routes that you have and
17:17I'm calling the first one S1 position
17:20and I'm setting it to zero and then I
17:22Define another one for the second one
17:24and my shoot long is going to be very
17:27simple I'm going to say return true if
17:30the S1 position value equals one that's
17:32it and we're choosing one as in a long
17:35position so that means for a short
17:38position we need to do the opposite so
17:40that would be this right if the S1
17:43position equals minus1 next I will take
17:46care of closing these positions and
17:48that's going to be using the update
17:50position method and in it I'm going to
17:53say if the value of S1 position equals
17:55zero I'm going to use the self.
17:58liquidate method to liquidate the
18:00current position for this trading route
18:02next I'm going to go to the second
18:03strategy and inside this shoong I'm
18:06going to say return if the shared vs
18:10S2 position equals 1 and we're going to
18:13do the opposite for a short position and
18:15again be careful with the key here it
18:17needs to be S2 because we're inside the
18:19second strategy and then I'm going to
18:23define the update position and do the
18:26opposite of what I did for the first
18:28strategy so I'm saying if the S2
18:30position equals zero that means we need
18:32it to be closed so we will liquidate the
18:34current position right so next we have
18:37the position sizing so I'm going to back
18:39to the first route and inside the goong
18:42method I'm going to say this the
18:44quantity of my position is going to be
18:46the margin one which is this value that
18:49I'm going to Define for the the size of
18:52the position of the first position and
18:54I'm using this utility function called
18:56size to quantity which will take this as
18:59the first parameter and then the current
19:01price because I want to open the
19:03position with the market order and then
19:04I'm setting the trading fees and then
19:06all I need to do is to say self. buy
19:08equals quantity and then the current
19:10price to open a position using the
19:12current price and for the short position
19:14I'm going to do the opposite so again
19:17quantity equals
19:19this but instead of self. buy I'm
19:22passing self. sell equals quantity and
19:25then the current price all right so
19:27let's go to the second route
19:29and do the same thing here so I'm going
19:31to say Quantity equals this except that
19:35instead of margin one it needs to be
19:36margin two and then I will say s. buy
19:40equals quantity and then the current
19:41price and then for the go short position
19:45we'll do the same thing again ensure
19:47that this one is margin two and then
19:49instead of self. buy we have self. sell
19:52all right I don't know what you're
19:53thinking right now but if it's not clear
19:55yet don't worry cuz it's going to be in
19:57a second but all you need to know is
19:59that at this point these two strategy
20:01files are connected to each other so
20:04wherever we change these four variables
20:06the S1 position S2 position and S1
20:09margin and S2 margin is going to take
20:11effect everywhere but now we need to
20:13define the logic of our strategy the one
20:16we just implemented inside this Jupiter
20:18notebook inside our first strategy so in
20:21fact I'm not going to touch the second
20:23strategy file at all anymore I'm
20:25completely done with it but for the
20:26first one let's begin all right so let's
20:28let's go back to the Jupiter notebook
20:31and here we were getting the candles
20:33like this using the research module but
20:35if I do it like this during a back test
20:38there's a good chance I'm going to use
20:39data from the future and we don't want
20:42that right because that would produce
20:43unrealistic results so instead I want to
20:46use Jess's built-in features for fishing
20:48the candles which insures we don't
20:50accidentally cause the liad bias in our
20:52strategy all right so I'm going to go
20:54back to the first strategy and I will
20:56Define a new property called C1 and in
20:59it I'm going to pass self. getet candles
21:02which is that built-in method for
21:04fishing the candles and again this one
21:06in Short you don't fish candles from the
21:08future Jesse will take care of it behind
21:10the scenes to only give you candles that
21:12you had at that point in time and the
21:14first parameter is the exchange and
21:16we're passing the current exchange and
21:18then the symbol now this one it could
21:21have Simply Be ET us right but because I
21:25wanted to Define it dynamically instead
21:27I I can use this self. rout and then I'm
21:30selecting the index of zero and then I'm
21:33selecting its symbol so this way no
21:35matter in back testing whichever symbol
21:38I choose here it will work here and I
21:40don't have to change anything so it's
21:42better now I'm also doing the same thing
21:44for time frame so I'm passing self. time
21:46frame which will give it whichever time
21:48frame that I pick here so instead of
21:50simply saying 15 minutes I'm using self.
21:54time frame and because I want only the
21:57closing price I as I showed you earlier
21:59in the jupyter notebooks I'm selecting
22:02all of them but only the closing prices
22:04which has the index of two but I'm also
22:06selecting only the last 200 candles and
22:09why is that well in the jupyter notebook
22:11we only had candles for 2 days right but
22:15when we are back testing let's say we
22:17want to back test it on multiple months
22:18if you give it so much data the mean of
22:21this data series is going to change a
22:23lot and it's not going to even be
22:25remotely what we need it to be so I want
22:27to make sure that we we catch the
22:29pattern of only the most recent data and
22:32that could be the last 2 days 3 days
22:34whatever but the point is it has to be a
22:38static number and I chose 200 you can
22:41try any other number and see whichever
22:43works best for you all right so now
22:46let's define another one and call it C2
22:50which is going to be very similar except
22:52we're selecting the second route by
22:54using the index one and everything else
22:56is the rest so you see inside my first
22:59strategy file I have access to the data
23:02or everything from even the second
23:04strategy so that's why I said I'm not
23:06going to touch the second strategy file
23:08at all all right now let's define the
23:11zcore right so first we need the
23:15spread and that is going to be self. C1
23:19minus self. C2 and again we're selecting
23:22from the index of one until the end
23:25which again is what I just did inside
23:26the Jupiter notebook and then I'm we
23:28going to say zscore equals utils zscore
23:32and I'm passing the spread and in the
23:34end I'm returning zscore minus one which
23:37will give me the last item inside this
23:40array and that's going to be the only
23:41one that I care about cuz it's not like
23:43I want to paint a chart with it right
23:45right so where do we Define the logic of
23:46the strategy or the rest of it now
23:48everything else will happen inside the
23:51before function now by the way in case
23:53you didn't know the before function is
23:55the first method inside the strategy
23:57cycle of J that we get called so
24:00whichever value that we set inside the
24:03before function is going to be
24:04accessible inside everything else inside
24:07the shoot long the go long the update
24:09position everything so that's why I'm
24:11doing all the decision makings inside
24:13this one and then setting the decisions
24:15inside these variables and then I'm
24:17checking for the result of those
24:19variables like for example in here all
24:21right now let's continue so once every
24:2424 hours or one day I want to
24:26recalculate everything like I want to
24:28see if the pairs are still Co integrated
24:30or things like that so I'm going to say
24:32this if self index equals 0 which means
24:37in the beginning or if the remainder of
24:40self. index divided by one day worth of
24:44one minute candles and we can get that
24:46by saying 24 multiped by 60 divided by
24:51this number here right now what is this
24:53number because we're using the 15
24:55minutes this is basically 15 but as it
24:59was the case with whichever whatever I
25:02did here so instead of hard coding I was
25:04using Dynamic values so that if we
25:06change something inside the js's
25:08dashboard it will take effect here right
25:11so because of that instead of Simply
25:13dividing it by 15 I'm going to divide it
25:16by this value I'm using the utility
25:17module again and the function time frame
25:20to 1 minute and then I'm passing the
25:22current time frame which is 15 minutes
25:24and it is returning 15 now I'm saying if
25:27the remainder of this division is zero
25:30that means it has passed 24 hours then I
25:34want to check for qution is qug grated
25:37equals utils do R cointegrated and then
25:40I'm passing the first set of price
25:41returns and then I'm passing the second
25:43one and if they are not going integrated
25:45anymore I want to close all the
25:47positions and I can do that by setting
25:49both these values S1 position and S2
25:52position to zero all right now let's
25:54move on next we're going to define the
25:56zscore and we can simply get that using
25:59the self. Z score which we defined in
26:01here right so let's get that one now I'm
26:04going to say if the current position is
26:06closed and we can simply get that using
26:09this and the
26:11zcore is below the threshold that we
26:15Define in here you remember these two
26:17lines it was
26:201.2 and minus 1.2 so I'm going to use
26:22that here so if the zcore is below minus
26:271 2 then I'm going to say go long with
26:31the first position by simply setting it
26:33to one and go short the second position
26:37by simply setting it to minus one now
26:39next I'm going to say let's remove this
26:43next I'm going to say else if it is
26:45already long but the Z score is above
26:48zero now in other words so you remember
26:51this we went long here but once it
26:54reaches the mean we want to close the
26:55position so I'm going to say if the Z is
26:58is now above zero set this one to zero
27:02and also this one all right so both
27:04positions need to be zero or closed and
27:08next I'm going to say if it's a short
27:09position and the Z score is below zero
27:12and that would be this case so if we
27:14went short but the mean is now below the
27:17mean then again close both positions and
27:20lastly I'm going to say else if it is
27:22closed but the zcore is above 1.2 which
27:26would be this case I want to short the
27:29first position and long the second
27:31position all right now we have the
27:32decision- making of our strategy we know
27:34when exactly it's going to go long and
27:37when is going to go short however we're
27:39still missing one variable and that is
27:41the size of the position and this is
27:43really important because the bigger the
27:45size of the market cap of an asset the
27:47less volatile it will be so in our case
27:50we already know that the price of
27:52ethereum will move less than the price
27:54of ethereum classic so let's say they
27:56are moving exactly at the same time if
27:58only the size of the positions are
28:00exactly the same for example if we long
28:02one of them with $1,000 and short the
28:04other one with $1,000 when they move
28:07let's say they both go up it's still the
28:09one with the lower market cap which
28:10would be ethereum classic in this case
28:12and that means if it was the long
28:14position we're going to be in profit but
28:16if it was the short position we would
28:17lose money right and that's not what we
28:19want so we need to make sure the size of
28:21these positions are not exactly the same
28:23value but they are somehow equal in
28:26matter of like how much money money we
28:28would make or lose between them right
28:30now how do we do that so first I'm going
28:32to define a new function and I'm going
28:35to call it set proper margin per
28:39route all right and I'm going to call
28:43this firstly in here and second in here
28:48so again this is where we opened the
28:51first long position and the second short
28:53position and this is where we did the
28:55opposite right and these two were we
28:57were just closing an already open
28:59position now this is incorrect going to
29:01remove it now to calculate this I'm
29:04going to use a formula which again
29:06luckily Jesse provides us so we're going
29:08to get the alpha but I'm not going to
29:10use the alpha so I'm just going to pass
29:11an underline and then the beta and is
29:15going to equal utils do calculate Alpha
29:18Beta And this is this formula so we're
29:20using the assets model again right and
29:25this is the formula so the x equals add
29:27constant then we're passing the second
29:29returns and we're using a model to do
29:32linear regression I think and then that
29:34will give us the Alpha and the beta and
29:36then we are passing them I'm not going
29:38to go through the exact math behind it
29:40again if you want to just watch that
29:42video from quantopian which they do
29:44explain this better now this is giving
29:46us both the Alpha and the beta to
29:49calculate the margin I'm using the
29:52current available margin multiplied by 1
29:55ided by 1 + beta and for the second
29:59margin or the margin of the second
30:01position or the second strategy I'm
30:03using it like this now to be honest with
30:05you I'm not a math wizard myself and I
30:08use chat GPT heavily for writing this
30:10part but because I did a lot of testing
30:12I know it's actually working so sorry if
30:14I cannot explain this any better all
30:16right so that's it now we have the
30:18margin one the margin 2 and the values
30:20for S1 position and S2 position and now
30:23we're ready to execute our first
30:24practice so let's go back to Jesse and
30:28we have both the trading routes eth Etc
30:32the 15 minutes time frame P trading is
30:35selected but the second one also needs
30:37to be press rating two and the duration
30:41make sure the fast mode is off because
30:43it actually cannot work with multiple
30:45routes yet so if I try to start it here
30:47I'm going to get an error so let's turn
30:49this off and I'm going to leave the
30:52actually let's turn off The Benchmark
30:54first and let's run it all right so this
30:58result isn't right and I went back and
31:01found the issue the issue is that we are
31:03passing the current candles but as it
31:05was the case in here we later converted
31:08the current candles into price returns
31:10so we need to do the same thing in here
31:12so I'm going to select this and instead
31:15return prizes to returns and then pass
31:18this and this all right and I'm also
31:21going to do it here all right so again
31:24we are converting the current candles
31:25into price returns and and that's the
31:28correct format for the calculations that
31:30we did with zcore and everything else
31:32all right so let's go back and run this
31:34one more
31:35time there we go this Equity curve looks
31:38unbelievably good and is for the last 12
31:42months the profit was 69% the Maxon isus
31:485% the M rate is 53% the sharp is
31:522.94 and the calmer is
31:5513.77 all right so let's go back and
31:58turn on The Benchmark feature too and
32:00run it one more time so while that's
32:03going I want to quickly remind you guys
32:04that I recently added support for Apex
32:06Omni which is a DEX which means you will
32:08keep the custody of your funds and
32:10there's no kyc you just connect your
32:12metamask wallet and you're good to go if
32:14you want to get started with Apex please
32:15consider using our link which will give
32:17you trading fee discounts and you'll
32:19also be supporting making these videos
32:21all right so check this out while the
32:23price was going up we made money while
32:26the price was going down we made money
32:29when it was in a Range we also made
32:31money so that's why we call this
32:33strategy a market neutral strategy cuz
32:35no matter if it's going up or down or in
32:38range you're going to make money as long
32:40as the bet that these two prize assets
32:43are going to Great is true right so
32:45that's why it's really fantastic so guys
32:48this result is actually too good to be
32:50true so I'm just going to buy a Lambo
32:52and make videos about how rich I am all
32:53right see you in the next video but hold
32:55on a second there's a catch multiple
32:57love them you see unlike my other videos
33:00or back test that I run I actually
33:03disable the trading fees in this case
33:05that's why we are not paying any trading
33:07fees and this is really important
33:09because it is taking
33:137,818 trades over a year that's simply
33:17too many trades and it's not like we own
33:19the exchange or are friends with them we
33:21are going to have to pay for trading
33:23fees and we're not even in hedge fund
33:25because hedge funds and I'm guessing
33:27Edward torb two they paid very little
33:29fees because they had special contracts
33:31but we don't so if we set the trading
33:33fee to what it actually would be on
33:35bance features and run this one more
33:39time let's disable the Benchmark
33:42feature now the result actually looks
33:45awful so we started by 10,000 and we
33:48ended with
33:50$343 now why is this is because again
33:53we're taking too many trades and we are
33:54paying just too much in trading fees
33:56we're paying 11 $1,000 in trading fees
33:59now it actually gets worse in reality
34:02not only we would have to pay this much
34:03trading fees but we're also going to
34:05have slippage and because in this
34:07strategy at least at the current state
34:10of it we're even counting on trades with
34:12very small margin of profit so that
34:14means if there's going to be slippage
34:16which there will be the results will
34:18look even worse than this so to get the
34:20real estate results we need to set this
34:23to a higher number such as this or maybe
34:25this one so what does this mean it means
34:27that if you're a H fund you can probably
34:28already trade this if you can get some
34:30kind of special contract or trade it on
34:32an exchange that has very little fees
34:34but for most of us that's not the case
34:36so what we can do is to implement as
34:40many filters as possible to make sure
34:42that we only take the trades that are
34:44worth it now the other thing that we can
34:46do is to trade only assets that are more
34:49volatile for example mem coins we know
34:51they are correlated with each other but
34:53they move much more and that means our
34:55trades probably won't have as little
34:57profit margin which would make it worth
35:00it for us I'm going to share the source
35:01code of this strategy on a GitHub
35:03repository and link to it down in the
35:05description it's going to be your
35:06homework to keep doing research and
35:09implementing filters to improve the
35:11results of this strategy and make it
35:13actually production ready you can also
35:15try it on other markets with different
35:16trading assets and see if it works there
35:18or preferably with exchanges that have
35:20lower trading fees I would love to see
35:22you guys on our Discord and share with
35:24me your experience with this strategy
35:26and let me know what works for you and
35:28what doesn't so that hopefully we can
35:30improve it if many of you guys want me
35:32to make a follow-up video about this
35:33strategy I might do it so just let me
35:35know about it now as always we're going
35:37to have a giveaway the random person who
35:38likes this video posts a comment and
35:40subscribes to the channel is going to
35:42win 1 million bunk token all right let's
35:43pick the winner for the previous
35:46video and the winner is rames thank you
35:49so much for your comment please reach
35:50out to me so that I can send you your
35:52bunk tokens thanks for watching and I'll
35:53see you in the next one
35:56[Music]
35:58oh
36:01[Music]

About this transcript

This page contains the full transcript of Pairs-trading strategy from scratch in Python by Algo-trading with Saleh, generated from the public captions YouTube serves with the video. The transcript has 6,924 words across 911 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.