Thanks for watching everyone! Yes, there are some simplifications regarding the buy limit order management. I was building this strategy with a more strict (and therefore more realistic approach) in another video. I was also avoiding looping using a vectorized approach. Check it out: ruclips.net/video/nAfnAuvRpqE/видео.html Cheers!
Thank you for your videos. I started programming trading bots a year ago. In the last semester I made a 'simple' AI trading bot as my final thesis. I am happy that I had your videos along me on this long jouney.
Thanks a lot Bajnok - that's awesome to read! Be invited to go more into details on your bot. Pretty sure a lot of people here would be interested in that as well (including me) :-)
Some thoughts, which might deterioate the performance: 1. the survivorship bias should be considered, i.e. when backtesting a system like that on the current index constituents, it is usually rated too good, because the bad performing stocks had been removed from the index 2. as a limit order already consumes margin, it is not possible to trade all 1000 of the Russell 1000 stocks. a sort of ranking will be needed (ATR works great on mean reversion systems) 3. in live trading you do not know, which limit orders will be filled at the beginning of the day. In backtesting, when checking for Low
Hey man, love the video. I have a question regarding one of you previous videos. For the python trading bot using sql it tells me that "" 'awai' outside function "". Do you know how I could fix that?
Thanks a lot mate. Sure, I have shown it here: ruclips.net/video/nQkaJ207xYI/видео.html And in the pinned comment here: ruclips.net/video/rc_Y6rdBqXM/видео.html
I haven't run the code so maybe I missed something but I think there is a bug: When you break after k > 10 in the while loop, you have no buydate and the first thing that you test for after the the break is: if buydate > last_selldate without a buydate I'm not sure what you wan't to do in this case. Maybe skip buying?
Yeah I didn't have time to cover that case anymore or it didn't come up during testing but this has to be covered indeed. That's happening because the Buy Limit order is not going through within the defined k. Just either stop the iteration or work with a try except here.
I agree. But I am not sure if BT allows you to exit the position after 10 time periods in backtest. If we changed that condition to something like 'take_profit = 0.05', then BT would be a better choice.
Hi Algovibes. Irgendwas stimmt mit den Loops nicht. Wenn ich folgendes eintrage: df = yf.download('HRB', start='2015-01-01') und den Python-Code mit einem frischen Kernel oder im Terminal ausführe erhalte ich die Fehlermeldung: NameError:"name 'buydate' is not defined. Did you mean: 'buydates'?". Woran kann das liegen?
Ja, das musst du noch covern. Das passiert, wenn die Buy Limit order nicht durchgeht. Entweder du brichst den loop ab oder arbeitest mit try und except. Hatte ich keine Zeit mehr zu covern. Video war lang genug :D
Hi dude, thank you for your work and expert tutorials. I have a small problem in that when i try to run the kernel i get the following error. NameError: name 'buydate' is not defined
Isn't there any python backtesting library out there that can handle more complex order types (buy limit, stop loss...) out of the box, other than the normal market order?
There is. And I have covered a bunch of them both in introductory videos and in implementation of a more complex strategy. Just check out the Python for Finance playlist. Thanks :-)
I haven't watched everything yet, but what if the Buy conditions are met after the low price was set ? You place an order at a price that may never occur in the next candles. Usually when backtesting you avoid using conditions on data from the current candle, especially with the high / low that are impossible to time. In a live market you will probably have radically different results. On of the safest way to avoid this kind of information bias is to build an event driven backtester or use the existing open-source ones ;)
The existing open source ones are VERY limiting. The fact that they are open source already says that there's also not a huge amount of money to be made with them either, or they would've never been put online for free. I looked through the top ones and pretty much came to the conclusing that the only way you can make a profitable one is to make one yourself tailored to your own strategies. If you can't trade profitable without a bot, chances are you're not gonna be profitable with one either.
@@wickedgummybear3104 You're not wrong that open source means that you won't have the same level of control as you would have with a custom engine. I would also add that some of these project have a pretty steep learning curve and you might just as well start from scratch. But you'll need to put some hard work (and rework) on your backtesting engine to make it somewhat close to real market conditions (spread, fees, etc...). When you say that they wouldn't be put online for free if you could make money with it, it sounds weird to me because that's just the dynamic of open source. It's just a framework and weither you succeed or not is just a question of how you use it. There are thousands of open source projects which are allowing people and companies to save/make money, some of them even find a fair, stable business model
First: I am your father! Second: The problem with the non filling Buy Limit Order is addressed in the video. Secondly please feel free to explore my channel. I have worked with a ton of Backtesting libraries already. These videos are made using as few libraries as possible and stay that way.
Love your videos. Keep them coming. From all the strategies that you tested until today what is the best (largest win rate and least drawdown)? May I suggest keeping a rank.
It is not clear. This is the only video I have seen him get truly excited about the outcomes. I have tried many of the strategies and have not found one that is working yet. Will keep looking!
@@Algovibes I was thinking about the broad concept of optimization. backtesting.py tries to maximize a given parameter, but sometimes (that's what I did with one of my bots) you have to clearly state that you want to maximize for example average profit while minimizing standard deviation, so a double objective optimization under constraint. I found pymoo package really useful.
why do i get different output, when i use the same code and same symbol (eurusd, 2015-01-01) ? without adding buydate= pd.to_datetime("1900-01-01") if doesnt even work
I do not understand the approach, why within the pandas data frame loop it is looked forward several times? I would take the orders and only look at whether the conditions are met or not and at most I would look backwards. This approach should be much simpler!?
Hi Franky1, I have shown different approaches to match buying and selling signals and decided for a loop logic here. You can use vectorized methods, mapping methods,.... The sky is the limit. Be kindly invited to check out all my previous videos in both the Python for Finance and cryptobot playlist. You will find a lot of concepts there! Cheers
Sorry, even though I installed this package, it still gives this error ModuleNotFoundError: No module named 'yfinance', maybe I didn't install it correctly because I'm new, or I don't know how to install api, please help me
Thanks for your videos. I have sent the mail for the activation of the subscription code but I did not have an answer when you can see it, I would appreciate it.
@algovibes you're not using the correct indicator as used by Rayner. He's using the "Stochastic" indicator, which is calculated based on the daily lows, highs and the most recent close. You're using the "Stochastic RSI' based on the close price only. These are NOT the same indicators.
Thanks Andre, you are right! Played a bit around with that and ends up in better results. Will see that I update. Anyhow I also played a bit around with intraday crypto data and find this strategy (taking stoch_rsi) to be interesting to form into a Trading Bot (no investment advice).
I haven't watched many of your videos but how do you and your community keep assessing strategies without using sharpe? Like for this video you have no idea what your risk adjusted returns vs buy and hold is. It's very, very naive to compare buy and hold to a strategy that gives 10 or even 5% of the returns. Do you take a strategy with 150% cagr but 90% drawdown vs a strategy that gives 50% cagr vs 11% drawdown? Of course not!
You have everything what you need to calculate the risk adjusted returns - so what are you waiting for? If you want to have a Backtesting output solution containing all important metrics (SR, DD, ...) you can also use a Backtesting framework as I did in many of the previous videos. Be invited to check out all (don't skip a single one and don't skip a second in every video) videos in the Python for Finance playlist (and additionally the cryptobot playlist).
@@Algovibes Thanks. I agree we have everything yes. Just surprised it wasn't part of your process here but as you say you have used Backtesting.py in your previous videos so it's not an issue. Thanks.
Please don't spend your time with these strategies, we all know that are bullish. The best idea is to detect liquidity (volume) in short time frame👍, please make us a video about this in Binance 🙏. Thank you for your channel 😎
He's not wasting time. There are alot of concepts/knowlegde in these video's that are helpful to alot of people. If you clicked on any of these video's hoping to find a 'get rich quick bot' then you're in the wrong place anyway.
Don't worry about my time guys. If you REALLY do consider becoming a channel member as time is money ;-) What I am doing on the channel is what I consider to be interesting.
Great video buddy!! Great content as ever. Just keep the good work going without any negativity affecting you. Request you to explain the .cumprod() with an example or give link of the video if ever you explained it. KUDOS 👍
When I try to run the code on the line: if buyprice>=df.iloc[row+k].Low: I get an error: single positional indexer is out-of-bounds Is this a typo on my end somewhere?
If I add some if's to prevent the Index Error the profits data at 23:22 looks the same, but if I change the ticker or the start date it differs from what you have. For example if I change the date like you did for the table at 24:02, the numbers become different. Did you change some additional settings in between these runs or is there a typo or something in the code that I use?
i suppose he will not answer to an old video. here here. had the same indexer-errors. and when using BTC-USD from 2015-01-01 i get different results... strange
![How To Build a Cryptocurrency Trading Bot with Python & the Binance API [SMA/RSI/Buy Limit Orders]](http://i.ytimg.com/vi/VZQg4KSmpQU/mqdefault.jpg)
![How To Build a Cryptocurrency Trading Bot with Python & the Binance API [SMA/RSI/Buy Limit Orders]](/img/tr.png)
![Backtesting Rayner Teos 1689% Bollinger Bands Trading Strategy in Python [Part II] - OVER 1K stocks](/img/1.gif)
Thanks for watching everyone!
Yes, there are some simplifications regarding the buy limit order management. I was building this strategy with a more strict (and therefore more realistic approach) in another video. I was also avoiding looping using a vectorized approach. Check it out:
ruclips.net/video/nAfnAuvRpqE/видео.html
Cheers!
Thank you for your videos. I started programming trading bots a year ago. In the last semester I made a 'simple' AI trading bot as my final thesis. I am happy that I had your videos along me on this long jouney.
Thanks a lot Bajnok - that's awesome to read!
Be invited to go more into details on your bot. Pretty sure a lot of people here would be interested in that as well (including me) :-)
Some thoughts, which might deterioate the performance:
1. the survivorship bias should be considered, i.e. when backtesting a system like that on the current index constituents, it is usually rated too good, because the bad performing stocks had been removed from the index
2. as a limit order already consumes margin, it is not possible to trade all 1000 of the Russell 1000 stocks. a sort of ranking will be needed (ATR works great on mean reversion systems)
3. in live trading you do not know, which limit orders will be filled at the beginning of the day. In backtesting, when checking for Low
Good points! Thx for sharing your thoughts Andreas.
very excited for the next video coming up
Awesome! Thanks for your support man.
Hey man, love the video. I have a question regarding one of you previous videos. For the python trading bot using sql it tells me that "" 'awai' outside function "". Do you know how I could fix that?
Thanks a lot mate. Sure, I have shown it here:
ruclips.net/video/nQkaJ207xYI/видео.html
And in the pinned comment here:
ruclips.net/video/rc_Y6rdBqXM/видео.html
@@Algovibes Thanks!
I haven't run the code so maybe I missed something but I think there is a bug:
When you break after k > 10 in the while loop, you have no buydate and
the first thing that you test for after the the break is: if buydate > last_selldate without a buydate
I'm not sure what you wan't to do in this case. Maybe skip buying?
buydate local variable used before it is assigned error is there
Yeah I didn't have time to cover that case anymore or it didn't come up during testing but this has to be covered indeed. That's happening because the Buy Limit order is not going through within the defined k. Just either stop the iteration or work with a try except here.
Yes, I had this issue
really interested in the implementation of this strategy for your next video! Looking forward to it!
Just released. Excited what you think about it!
Your code is more beautiful than the strat itself.
I am wondering if this could be implemented using backtesting.py library. Is it make sense to try? Or too much complicated?
I agree. But I am not sure if BT allows you to exit the position after 10 time periods in backtest.
If we changed that condition to something like 'take_profit = 0.05', then BT would be a better choice.
Interesting question and yes, that would be possible as well!
Hi Algovibes. Irgendwas stimmt mit den Loops nicht. Wenn ich folgendes eintrage: df = yf.download('HRB', start='2015-01-01') und den Python-Code mit einem frischen Kernel oder im Terminal ausführe erhalte ich die Fehlermeldung: NameError:"name 'buydate' is not defined. Did you mean: 'buydates'?". Woran kann das liegen?
Ja, das musst du noch covern. Das passiert, wenn die Buy Limit order nicht durchgeht. Entweder du brichst den loop ab oder arbeitest mit try und except. Hatte ich keine Zeit mehr zu covern. Video war lang genug :D
Hi dude, thank you for your work and expert tutorials. I have a small problem in that when i try to run the kernel i get the following error.
NameError: name 'buydate' is not defined
Very welcome man! Well you need to define the buydate somewhere. Where exactly are you running into this? (timestamp would be nice)
This is great i'd love to see if this works on live markets like Crypto, FX, Indexes and on lower timeframes. Could be very interesting
Thanks for your feedback man :-)
Isn't there any python backtesting library out there that can handle more complex order types (buy limit, stop loss...) out of the box, other than the normal market order?
There is. And I have covered a bunch of them both in introductory videos and in implementation of a more complex strategy. Just check out the Python for Finance playlist. Thanks :-)
@@Algovibes so let's stay specific please, which libraries support these order types?
hello, can you please tell me how to fix this error: " AttributeError: module 'ta' has no attribute 'tredn' "?
I actually can: Just fix your typo :-)
Hey, can you upload the function version of the same code and upload it on the drive, and can you tell me how I can access the drive?
Hi Aman,
sorry did you already drop me a mail?
I haven't watched everything yet, but what if the Buy conditions are met after the low price was set ? You place an order at a price that may never occur in the next candles.
Usually when backtesting you avoid using conditions on data from the current candle, especially with the high / low that are impossible to time. In a live market you will probably have radically different results.
On of the safest way to avoid this kind of information bias is to build an event driven backtester or use the existing open-source ones ;)
The existing open source ones are VERY limiting. The fact that they are open source already says that there's also not a huge amount of money to be made with them either, or they would've never been put online for free. I looked through the top ones and pretty much came to the conclusing that the only way you can make a profitable one is to make one yourself tailored to your own strategies. If you can't trade profitable without a bot, chances are you're not gonna be profitable with one either.
@@wickedgummybear3104 You're not wrong that open source means that you won't have the same level of control as you would have with a custom engine. I would also add that some of these project have a pretty steep learning curve and you might just as well start from scratch. But you'll need to put some hard work (and rework) on your backtesting engine to make it somewhat close to real market conditions (spread, fees, etc...).
When you say that they wouldn't be put online for free if you could make money with it, it sounds weird to me because that's just the dynamic of open source. It's just a framework and weither you succeed or not is just a question of how you use it. There are thousands of open source projects which are allowing people and companies to save/make money, some of them even find a fair, stable business model
First: I am your father!
Second: The problem with the non filling Buy Limit Order is addressed in the video. Secondly please feel free to explore my channel. I have worked with a ton of Backtesting libraries already.
These videos are made using as few libraries as possible and stay that way.
Love your videos. Keep them coming.
From all the strategies that you tested until today what is the best (largest win rate and least drawdown)?
May I suggest keeping a rank.
It is not clear. This is the only video I have seen him get truly excited about the outcomes. I have tried many of the strategies and have not found one that is working yet. Will keep looking!
function logic /OOP version would be great to see. I spent two hours slowly working through this. Very helpful as usual!
Great video ! I am wondering if instead of maximizing as backtesting.py does it is a way to minimize losses (tried pymoo but hell slow 😅)
Thanks mate. Can you elaborate on what you tried or would be interested in? Didn't quite get it, sorry!
@@Algovibes I was thinking about the broad concept of optimization. backtesting.py tries to maximize a given parameter, but sometimes (that's what I did with one of my bots) you have to clearly state that you want to maximize for example average profit while minimizing standard deviation, so a double objective optimization under constraint. I found pymoo package really useful.
so rainer use simple stoch? you use stochRSI?
You are right!
why do i get different output, when i use the same code and same symbol (eurusd, 2015-01-01) ?
without adding buydate= pd.to_datetime("1900-01-01") if doesnt even work
Have you made any videos for trading on pancake swap ? This would be very interesting
Not yet
You are the best man! Never stop these videos 🔥
Never! :-) Thanks for your support man
please explain the code, how to code to reOrder_long when the price is bullish or trend up. I want first order 5% second order 20%, and so on
How Can I get such strategies or my strategy coded to trade automatically.
I have a whole playlist on topics like that. Be invited to check that out! (Cryptobot playlist)
Please let me know how can I buy this strategies from you to use for algo trading .
Nice video, as always I have learn something new every video.
Awesome. Thanks a lot!
can you make videos on trading with leverage and Ichimoku indicator ?
Hi Salem, thx for your suggestion. Will see what I can do!
@@Algovibes Thank you so much my dear freind !!
Wonderful content as always. Please run it on ran-time crypto market. I think we all want to see the result.
Would be interested in that as well! Thanks for your feedback.
I do not understand the approach, why within the pandas data frame loop it is looked forward several times? I would take the orders and only look at whether the conditions are met or not and at most I would look backwards. This approach should be much simpler!?
Hi Franky1, I have shown different approaches to match buying and selling signals and decided for a loop logic here. You can use vectorized methods, mapping methods,.... The sky is the limit. Be kindly invited to check out all my previous videos in both the Python for Finance and cryptobot playlist. You will find a lot of concepts there! Cheers
How to apply this code in live market, pls make video on this...
Just released the video. Looking forward to your thoughts on it!
@@Algovibes Thanks bro 👍
Can I monitor live stocks using python ?
Sure! Just need an API providing live stream of stock price data.
Does it work by prediction or not?
Can you elaborate on your question?
For a second I thought that was you in the beginning of the video, a German/Asian, would have been surprising
Will do face reveal at 100.000 subscribers (with a pinch of luck probably in the next 2-3 years)
Thanks for your content.
Could you give me an idea to make a bot for binance with dynamic stop loss? 😊
Welcome mate,
that could be helpful as I was playing around with trading stop loss here:
ruclips.net/video/V6z1ME3-0_I/видео.html
Sorry, even though I installed this package, it still gives this error
ModuleNotFoundError: No module named 'yfinance', maybe I didn't install it correctly because I'm new, or I don't know how to install api, please help me
If it doesn't find the module your current running python version is probably not added to your windows path..
You need to install the module first :-) in Jupyter just type in: !pip install yfinance
Thanks for your videos. I have sent the mail for the activation of the subscription code but I did not have an answer when you can see it, I would appreciate it.
Added you! Thanks a ton for your support.
@algovibes you're not using the correct indicator as used by Rayner. He's using the "Stochastic" indicator, which is calculated based on the daily lows, highs and the most recent close. You're using the "Stochastic RSI' based on the close price only. These are NOT the same indicators.
So it should be "ta.momentum.stoch" instead?
@@franky12 yupp! And then first three parameters in order are high, low and close. Then window (set to 10).
Thanks Andre, you are right! Played a bit around with that and ends up in better results. Will see that I update.
Anyhow I also played a bit around with intraday crypto data and find this strategy (taking stoch_rsi) to be interesting to form into a Trading Bot (no investment advice).
Thank you for your code it is remarkable as usual. Looking forward for the bot version
Thanks a lot Victor. Will surely come!
you scroll and rush through the results too fast! i mean , its like the most important part of the video.
Thanks for the feedback Fred. Just pause the video if anything is going to fast or play the video at a lower speed.
I haven't watched many of your videos but how do you and your community keep assessing strategies without using sharpe? Like for this video you have no idea what your risk adjusted returns vs buy and hold is. It's very, very naive to compare buy and hold to a strategy that gives 10 or even 5% of the returns. Do you take a strategy with 150% cagr but 90% drawdown vs a strategy that gives 50% cagr vs 11% drawdown? Of course not!
You have everything what you need to calculate the risk adjusted returns - so what are you waiting for?
If you want to have a Backtesting output solution containing all important metrics (SR, DD, ...) you can also use a Backtesting framework as I did in many of the previous videos. Be invited to check out all (don't skip a single one and don't skip a second in every video) videos in the Python for Finance playlist (and additionally the cryptobot playlist).
@@Algovibes Thanks. I agree we have everything yes. Just surprised it wasn't part of your process here but as you say you have used Backtesting.py in your previous videos so it's not an issue. Thanks.
@@Algovibes I'm painfully trying to makt this into a class and use the backtesting library lol.
@xaknafein try this:
profits = pd.DataFrame([(sell - buy)/buy for sell, buy in zip(sells,buys)])
profit_vol = profits.std()
profit_mean = profits.mean()
sharpe = profit_mean/profit_vol
# Vectors for cumulative Product and buy_Hold
cumprod = (1 + profits).cumprod()
buy_hold = (df.Close.pct_change() + 1).cumprod()
#Cumulative Return Values
cumprod.iloc[-1].values
buy_hold[-1]
algo_results = {"Results": "Reyner Teos Stochastic Model", 'Average Return': profit_mean,
'Volatility' : profit_vol, "Sharpe": sharpe, "Cumulative Return": cumprod.iloc[-1].values,
'Buy & Hold Return': buy_hold[-1]}
results_RSI_BNB = pd.DataFrame(algo_results).T
Waiting for the bot
Awesome. Will surely come!
Thanks magister
Thanks for watching mate
Please don't spend your time with these strategies, we all know that are bullish.
The best idea is to detect liquidity (volume) in short time frame👍, please make us a video about this in Binance 🙏.
Thank you for your channel 😎
He's not wasting time. There are alot of concepts/knowlegde in these video's that are helpful to alot of people. If you clicked on any of these video's hoping to find a 'get rich quick bot' then you're in the wrong place anyway.
Don't worry about my time guys. If you REALLY do consider becoming a channel member as time is money ;-)
What I am doing on the channel is what I consider to be interesting.
lets run this on 1000
Bro accidentally found a billion dollar forex strategy
lol :D Not really.
3588% ? 😅 It seems absurd
Was just quoting Rayner 😛 And to be fair it is over 22 years.
cheater
?
Great video buddy!! Great content as ever. Just keep the good work going without any negativity affecting you. Request you to explain the .cumprod() with an example or give link of the video if ever you explained it. KUDOS 👍
Thanks my friend! Sure. Explained return calculation here:
ruclips.net/video/fWHQwqT3lNY/видео.html
When I try to run the code on the line:
if buyprice>=df.iloc[row+k].Low:
I get an error:
single positional indexer is out-of-bounds
Is this a typo on my end somewhere?
If I add some if's to prevent the Index Error the profits data at 23:22 looks the same, but if I change the ticker or the start date it differs from what you have. For example if I change the date like you did for the table at 24:02, the numbers become different. Did you change some additional settings in between these runs or is there a typo or something in the code that I use?
i suppose he will not answer to an old video.
here here. had the same indexer-errors. and when using BTC-USD from 2015-01-01 i get different results... strange