hi and welcome. Today, we're going to review how to trade the price Channel breakouts by automating a new indicator in Python. The algorithm will be able to detect price channels, and when the breakout happens, it will generate a trading signal in the direction of the breakout. Then, we will backtest this indicator in a strategy, showing also a smart trade management variation that will increase our backtesting returns from 20% up to almost 32%. So, the purpose here is not only to show you an algorithmic trading approach but also to give you powerful ideas that you can introduce into your classic manual trading as well. But first, we're going to prove these ideas numerically through backtesting to make sure that they really work. So, as you can see, it's a good-looking approach. I will take you through the details of the strategy itself, and then we'll do the coding and the backtesting in Python. If it's your first time on this channel, the Python source code is available from the link in the description below as usual. So you

can download it for your own experiments. The indicator is built as follows. First, we need to look for pivot points, and these are candles with extreme price values where the price is either greater or lower than all the neighboring candles from both sides. These are the purple points that we can see on this chart, and they represent pivots detected by our algorithm. So, we can see that these are indeed extreme values, and the algorithm is providing good results. The second step is to fit these points in linear slopes, and then the third step is to define a set of conditions to detect a valid price channel. So, what is a valid price channel? For these conditions, first, we need to have a minimum of five pivot points. These can be two high pivots and three low pivots, or also the opposite, three high pivots and two low pivots. We can also have more than five points in total, but five is the minimum to start detecting a channel, so six is okay as well. Then, when we apply the fitting slopes, the R-squared goodness of fit values should

be greater than a threshold, which in this example, I'm taking as 0.9. If you're unfamiliar with the R-squared coefficient, it essentially shows the proximity of the points we're attempting to fit to the slopes that are going through these points. So, we need this coefficient to be as close to one as possible for both high and low pivot fitting. The third condition is that the difference between high and low slopes should be less than a threshold, for example, around 10% in this example, which means that the slopes are almost parallel. Remember that a price channel should have those slopes almost parallel; otherwise, we're going to fall into different patterns such as triangles, ascending triangles, descending triangles, flags, and so on. And this is how we detect a valid price channel numerically. Next, we need to detect the breakout. So, whenever we have a channel, we will wait for a candle that opens and closes either below or above the channel. In this example, we're seeing a bearish candle whose open and close prices are both below the lower slope of

the channel, so we can confirm it's a breakout signal. The opposite direction, if we have a candle that opens and closes above the upper line of the channel, then we have a breakout in the positive direction, and this is how the signal is generated in both directions. Now we need to put all of it together to make a sample strategy that we can backtest. Let's take this bearish breakout as an example. Since it's a bearish breakout, we will only consider opening a short position, but we will not do this immediately when the signal is detected. Instead, we will put a sell order a bit above the current closing price where the signal is occurring, and by a bit, I mean within a few percent of the current price. It's around this green line that we can see in this example. Remember, this is just an order, so it's not an open position yet. This way, if the price retraces back to the order level, it will trigger our position on the market, and this is one of the main ideas I need to

verify and show you in this video. The reason we do this is because we expect the price to retrace after a sudden breakout before taking its normal trend direction. So, whenever we have a breakout, we expect the price to retrace back to a certain level before carrying out with a new trend. And remember that we are trying to squeeze a few percents more into our favor in each trade we're going to open. So, we're fighting the market and we need to use all the tools at our disposal, even if it's just a few percents into our favor. Then we can define the stop-loss, which is also fixed to a distance within a few percents of the current price, and the take profit is obtained using the stop-loss distance multiplied by the take profit stop-loss ratio. So, in summary, we have three parameters defining our trade management system: percentage of the order, percentage of the stop-loss distance, and the take profit stop-loss ratio. We will code all of this in Python. We will also include a few variations using the RSI for

example, and we will see how it all affects our trading system. And again, the whole code is available for download from the link shared below in the description. So good luck and see you in the Jupiter notebook. This is our Jupiter notebook file. I'm importing whatever we're going to need as libraries. We're reading the CSV data file, so EUR/USD candles, the daily timeframe between 2003 up to 2023. So this gives us roughly 20 years of data. And then we are using a function that detects the pivot points. So these are the purple points that we have seen in the presentation at the beginning of the video. These are the pivots high and the pivots low. So this function takes the candle index, then a window, which is the number of neighbors to consider on the left and on the right of the candle before testing if it's a pivot high or a pivot low. And the function returns an integer, either one for a pivot high, two for a pivot low, and three if you have both like a candle.

that is has extreme values, so it's higher and lower at the same time of all of its neighbors, which is a very rare occurrence, but we need to categorize it as the other categories as well. Then, then we can Define another function called collect Channel. It takes the candle index, it takes the number of back candles, it takes a window, and a parameter called parallel. So, this is the percentage that uh, the percentage threshold to check if the slopes of the high and low pivots are parallel. And as you can see somewhere around here when we are using the function, we're using parallel equal to 0.1, which is 10%. So, parallel is equal to 0.1, this means 10%, just as we have mentioned for this example. So, again, the function takes the index of the candle to be tested, the current candle. So, it's going to test for the current candle if it's preceded by a channel, a price Channel, and the number of candles that we're going to consider before the current candle, just a sliced window of the

previous candles to detect if there was a channel is limited by this amount. So the number of back candles usually I'm taking 40 as you can see here. So for each candle I'm considering the previous 40 candles to test if we have a channel and if we have any breakout signal around it. So then we have a window, this is a gap window that we're going to leave. It's to avoid uh look ahead bias so whatever we're doing we don't need to look into the future and since we are using the is Pivot candle that needs to read on the left and on the right so on in the future of the candle we can't read really in the future so we're going to shift all of our stuff study here by a certain amount which is called window so this is why window was introduced to make sure that our computations are still correct are still realistic and window should always be greater than whatever you use for the pivot here window so if we've used three here it should be minimum of three if we've used

four for the pivots here we should use five. For example, or at least four. So just to avoid looking into the future because this wouldn't be correct anymore. It wouldn't be realistic. Inside of this function, we're taking a local data frame. So the slice of the candles that we're going to apply our study on. We're checking the is Pivot. So we're applying is Pivot. And then we are looking for highs, the pivot highs. I'm taking only the last three of those. If we have three. So if it's two, then we're just taking two. But we're limiting the slice to the last three points. Same, we're storing also the indexes of the highs, the lows as well. And they're in indexes. Remember, we're taking maximum three lows and three highs at this point. And then we're checking the total length. So how many points in total do we have? If the total length is above five. So this is the condition we already talked about. We need at least five pivot points to collect a channel. And this is the

condition here. Then we need also at least two lows and at least two Highs at the same time. So we don't want to have, for example, four highs and one low low, because in this case this condition is verified, but we can't construct a channel. So these are the conditions to be able to fit uh points into a channel. And then we are going to use the linear regression from stats uh providing the lows, the indexes, and um Val values of the lows. And also the highs and their indexes. So this is how we fit the pivots into two slopes. Then we're going to apply the conditions on the slopes to make sure they are almost parallel within a few percentage, which is the uh parameter value here called parallel. Now, if we provide parallel equal zero, then we are discarding this condition. We will take any kind of Channel, could it be a triangular Channel, ascending triangle, descending triangle, and so on. But as long as you provide a positive value for the parallel, we are sure that we are only selecting the slopes that are a

bit parallel. Let's say within this provided threshold. So as you can see, this function is going to be very versatile. You can copy it and paste it in your different programs. You don't have to use it only in this Jupiter notebook. You can take it as is and make sure to adapt it for your different strategies as well. Okay, so now we've defined the back candles equal 40. Our windows equal to 4. We're going to define a new function called Point position. And this is just for visualization purposes. So whenever we have a pivot, we're going to add a point above or below the candle, depending if it's a pivot high or a pivot low. And this is just to be able to visualize these points right here. So I'm defining this function right here. We're applying it to the data frame after computing the pivots. Wherever we have a pivot, we're going to place a point, an additional point. We're going to provide a candle like a central candle of interest. For example, slice around the

candle. Then we're going to plot the candle sticks around it, and we're going to add using the ad scatter the point positions wherever we have pivots, and we're also computing the signal at this point just to make sure that if we have a signal, we are computing, or we are printing it right here. So at the moment for this candle, we don't have a signal, and this is what we have. So at this point, we can plot any candle or slice around the candle that we are interested in. So this is candle 422. We can go right here. So where is 400, uh, 422 is somewhere around here. We don't have any signal here, so there's no breakout to be considered. It's just that we're slicing, we're plotting the pivot points, and we're also plotting the future to know what happened after in case we have a signal. So now we reach the important part of our algorithm. So we're going to detect the breakout when the breakout happens. So first of all, we are defining this function is

breakout. It takes a candle index, so if the candle is a breakout, it's going to return the signal, an integer which notifies us about the direction of the breakout. So, return one if it's a bearish one, return two if it's a bullish breakout. Then we consider the back candles and the window parameter, just as we have explained these previously, and the parallel as well. So, this is just an edge case. Then we are defining these parameters here. So, we're going to collect the channel and the collect Channel function that we have previously defined. This one is not only returning a signal, it's going to return a lot of parameters. It's going to return the slope of the lows, the slope of the highs, and many different parameters. And these are saved here in these variables. So, we have the slope of the lows, the intercept of the lows, the slope of the highs, the intercept of the highs, and the r square of the lows fitting, and the r square of the highs as well. So, we're

going to need all of these parameters in our function. Then we're going to define a set of variables here. So, the previous index is the index of the previous candle. Remember we are applying this function on a specific candle, so previous index is candle minus one. The previous high is the high of the previous candle, the previous low as well, it's the low of the previous candle, the previous close, and so on. And then we have the current index, the current high, current low, current close, and current open. We're going to use these to define different variations of our breakout because you might want to experiment with these. So for example, the current setup is the following: if the current open or the current candle is opening below the lower slope of the channel, and at the same time the close, it's closing also below the channel. So the whole candle opened and closed below the lowest line of our channel. And at the same time the R-squared of the high and the low are 0.9, both of them, then we are returning one. So this is important to

consider so it's a candle that is opening and closing below the current um Channel. But you might want to experiment differently. You might want to say uh it's okay if it opens within the channel. Then you might want to discard this condition. You just want it to close below the channel. Okay so you're a bit more flexible in this case less selective I would say. So we would just keep one condition. You might want to add the high and the lows as well. You might want to consider the previous candle conditions as well. So it depends how you want to detect the breakout. In this case I'm taking a simple case where if the current candle opens and closes below or above the channel then we have a signal. But at the same time we need a good fit so uh the slopes are fitting within 0.9%. So this this is good. And of course the parallel setup of the highs and the low slopes was introduced here. This is a condition that was introduced used in collect channel. So it's it's right here.

so anyway, we wouldn't have any returned values if we wouldn't have a parallel Channel, and that's it, basically. So if we have a signal, we return one or two depending on the direction of the signal. Otherwise, we return zero. So this is the breakout function. Now we can apply it to our data frame. So again, I'm defining window equal 5 for now, back candles equal 40, and these are two parameters that you might want to play with and experiment on. They might change and affect our final results. Anyway, for the sake of this video, just as an example, I'm defining a new column in our data frame called is breakout, and I'm applying the is breakout function with the necessary parameters on all the data frame, and this is how we have a new column now called is breakout, which will hold the signal, as you can see here. So if it's a bearish or a bullish signal, one or two. And here, I'm print all the candles where I have a signal that is

different than zero that we will be using later on so we can visualize these signals as well. So I'm looking for, for example, for [Music] 1,840 like this candle. Let's check it out. So, this sale is just going to plot the candles, plot the pivot Points, and then plot the uh signals or the breakout uh points as well. Let's take a look at this. So we have this one, the signals are here. They are consecutive signals, but it's basically the same signal. We're just going to take the earliest one 'cause we need to know as early as possible what's happening. So we're not going to wait for more signals. And as you can see, we have a channel going down and it breaks out. And then the signal is triggered here. We're going to put an order a bit lower within a few percentage of the current close. And this is going to be somewhere around here. If you're taking, I don't know, like 3% or 2%, it should be somewhere around here.

so that's good. So we're opening here, and then we're waiting for the price to climb back up. We could also try another candle. For example, let's take the first one, 109. So I'm going to change this here. So we on this, and then we have also a down channel, as you can see. And then we have a breakout right here, and this one is an excellent signal. So we just had a breakout, and if we put the order a few percents less, we're going to catch a very nice uptrend, as you can see here. So provided that we know when to exit this trend because then it's not infinite, but it looks fine for me from this perspective. Remember that this is the daily time frame, so even if you left it for a month, it's going to be a month of wins. So this is a great signal. Let's check another one, actually. So these are 109, 110, 111, 12, 13, 14, up to 16. This is the same signal, as you can see here.

these are the stars that are consecutive. So, these are the consecutive signals happening right here. So, this is one signal. Let's check another cluster of signals. Let's say 564, for example. So, 564, same here. We have a down Channel. Then we had a breakout in the positive direction. We didn't have a very long uptrend. So, as we can see, it was very limited. But provided that you have a good trade management system, we shouldn't incur a lot of losses in this case. But anyway, so the algorithm is working well. We don't know on the long term the strategy if it's enough on its own. But we can run a back test on this. So, to back test the strategy, I'm using the back testing Library. We're providing the size of the um of the trades positions. I'm taking a take-profit stop-loss ratio of 1.5 percent 0.03, as you can see here. And that's the uh stop-loss percentage. So, I'm not using the order. We're entering the market.

directly when the signal is occurring, we're using a margin of one over five, so it's a leverage 1 to 5, and a cash of $10,000 as a start going to run this part. So first let's compute the RSI because we're going to use it later on. So RSI length 4, remember this is the daily time frame, um, defining this function of the signal that's just returning the breakout column from the data frame, and then we can start our back test, and we have a return of 19%, a win rate of 46%. Remember that we have a 1.5 ratio to take profit stop loss ratio, so 1.5 should be fine for 46% win rate, and we have a drawdown of -15%. That's a lot. I wouldn't recommend anything below 10%, and an average drawdown of -1.39%, so that's okay, that's a good one, but the maximum was a bit scary, so that's this is the worst.

case of our drawdowns and we can plot this actually going try to plot this. We have an increasing Equity, we have drawdowns periods as you can see, but we have an increasing equity and we end up with 19% of returns. Some of the trades are good are happening at the correct time, some of them are being stopped. So now we're going to introduce another percentage which is the order percentage, and we're going to separate these two variables. So we have the stop-loss percentage and the order percentage. The percentage of the order is 0.01, so I'm taking 1% instead of opening directly the position. I will be putting an order below or above the current signal by 1%. And then the stop-loss percentage is set to 3% for the stop-loss distance, and I'm adding another way to exit the trades is when if we have a long position and the RSI climbs above 95, then we're going to

close immediately because we have an overbo asset. And the same thing in the opposite direction if we have an open position short position and the RSI drops below five then we also close, even if we didn't reach the stop loss or the take profit values yet. So we're going to use the RSI as another way to exit the strategy or to exit our trades. And this time I'm using a take profit stop loss ratio of 1.8 percentage 0.01 for the order. I could do this because we're using the percentage. Remember that we are not entering the market directly, we're squeezing a bit more to our advantage into our favor from the market, waiting for the retracement. And this is why I could immediately increase the take profit stop loss ratio at this point because I want to put into our favor from both directions the point from the point we are opening the trade. This is why we're using the order and also from the point we are closing the trade. And this is where the take profit

stop-loss ratio comes in so both directions. The both points, the entry and the exit, are getting wider if you know what I mean. And then the stop-loss percentage is 0.03. So now we can backtest this one. I'm going to run this backtest, and we're going to see that we have 32% in returns here. So we've improved our um, our results using the same indicator just by modifying the way we are uh trading, we are entering and exiting the trades. This is to show you the importance also of the trade management part in trading. If we have climbed from 20% up to 30%, that's one-third of our returns. So before we had around 20% for the same indicator, 19%, and now we have 32%. That's 13 of our returns more. And this is just by changing the trade management approach. So I think we should also emphasize on this part as much as we are concentrating and focusing on forecasting the price direction and the

good trading indicators as well now we can plot whatever we have with the um and we can see that we have um. We have an increasing Equity so we have some drow down zones as you can see here but overall it's an increasing uh Direction so this is over the Euro US dollar which is fluctuating so it's not a trendy Market. It's fluctuating in all directions so it's not biased towards buying or selling it's using both directions in trading and this is it for this video. I hope you guys liked it. I hope you found the information helpful until our next one trade safe and see you next time

Boost Trading Returns by 30% with Python Automation of Price Channel Breakouts

Education channel providing simple and quick lessons on related topics: Python, Data Analysis, Machine Learning and data driven applications like stocks and currencies trading. 
I hope these videos will be of help, Good luck to you in your coding journey !

🍓Discover how easy Python can be. Learning Courses Discounts (links below):
* Algorithmic Trading In Python
* Python basics, 
* Object Oriented Programming In Python
* Data Analysis In Pandas And NumPy
* Machine Learning In Trading

Python is easy to learn and if I could do it then surely you can as well !

DISCLAIMER: The information contained in this channel is presented for educational purposes ONLY, I am NOT a financial expert nor a financial consultant, if you decide to use any of the content of this channel you may do so at your own responsibility. Although I do my best to provide reliable information I am NOT responsible for any losses incurred due to any trading systems presented on this channel.


Boost Trading Returns by 30% with Python Automation of Price Channel Breakouts

Boost Trading Returns by 30% with Python Automation of Price Channel Breakouts

스크립트