If you want to learn algorithmic trading, check out this free course that I created here on UA-cam: ua-cam.com/play/PLtqRgJ_TIq8Y6YG8G-ETIFW_36mvxMLad.html
What helped me tremendously was creating a playback engine in Python that uses recorded data to perform a "walkthrough", but without the restrictions of real time. So I record 1 sec data, and then play it back by reproducing the data flow as though it was real time. With this method, you can replay a full day as fast as your computer can rip through it. For me it is about 1-3mins depending.
Thanks for the comment. I currently don’t create algorithms for anyone else. But if you want to learn how to code your own options trading bot, you could check out this video in which I code an example options trading bot: ua-cam.com/video/Lq-Ri7YU5fU/v-deo.html
Former professional tester here. I was writing up the same comment. The "backtesting" talked about in this video seems rather badly broken - or possibly it is intentionally crippled in order to save CPU cycles in the cloud instance.
I'm happy I developed a trading strategy for what I want my algo to do before I even started learning. Learned alot from this. I just started my algo trading journey and appreciate all your videos
Walk Forward testing is a technique where you backtest with staggered periods of optimised and out of sample data. I think what the OP meant to say is Forward Testing.
This is not what walkforward testing is! This is live (paper) trading testing. Walkforward testing is backtesting with a sliding window. Other mistakes to avoid: - relying on a single strategy vs a portfolio of not-too-correlated strategies - not monitoring the live strategies
Hi Jul, Thank you very much for your comment and mentioning other common mistakes. Paper trading with live data is a special case of walkforward testing. I should probably have specified that more clearly. Thanks for pointing that out.
I would disagree that walkforward testing is a must. Instead I start out small but with real money. Every loss hurts and you analyze the trades much more. It also motivates you to continue when you earn real money.
Hello. This was a very good expose on the pitfalls of algorithmic trading. Here's my question: what if one's broker doesn't charge any commission and/or trading costs? What are costs that are still left to consider? Thank you in advance
Thanks for the comment. Often there are clearing fees that aren't mentioned by the brokers since they technically aren't broker commissions. But even with no clearing or broker fees, there are other costs. Slippage is one of the biggest non-obvious costs. Slippage is the money you indirectly pay by getting a worse fill price than your initial order price. Otherwise, there also are market impact costs. Depending on your trade size, and the liquidity, your orders can directly move the price of the security. This market impact will usually lead to worse entry prices. Depending on what you trade, you might also sometimes not be able to get filled at all. If, for instance, a security is hard-to-borrow, you might not be able to locate any shares to short. Or you might have to pay high margin rates/borrow fees. I hope this helps. If you have any follow-up questions, definitely let me know.
@@TradeOptionsWithMe This seems to assume that the trader is consistently working with even lots and several lots at one time. But what about smaller trade sizes? I'm thinking maybe 100 shares or 200 shares. And does 'slippage' occur with a market order or a limit order?
@@William_sJazzLoft It really depends on what you are trading. If you are trading highly liquid, high-float equities with a few hundred shares, your market impact costs likely are quite negligible. But if you are trading iliquid, low-float penny stocks even a few hundred shares can have an impact. Negative slippage can usually be prevented by using limit orders, but this comes with the risk of not getting filled which you can think of as an opportunity cost.
@@TradeOptionsWithMe OK. Although I don't generally deal with penny stocks, I do for the moment generally trade in stocks that are priced under $10. I have made remarkable progress having done so thus far. I'm on track to clear a cumulative profit of 10% in this next trading week. But I use limit orders almost exclusively ( both buying and selling ). I may have to be a little bit more patient - but I always get filled. And that was one reason I asked you the question about what cost might still be incurred. But still I appreciate the admonitions and advise that you are giving
Slippage and the spread are the largest trading costs that I have encountered. The video author has already covered the topic of slippage quite well. The spread can be quite large in certain cases; for example, going long with the ETF TQQQ reveals fairly small spreads due to good liquidity, but SQQQ has much larger spreads because it trades to the closest penny but the stock was only about $5 a share at one point. A recent 5 for 1 split helped this problem, but after a market rally the spread problem is starting to appear again (and cut into profits). Your algorithm has to, on the average, make more profit per trade than the spread. This is easier said than done, especially for day trading algorithms that trade frequently. I found that the spread was a much greater obstacle than trading commissions, since it scales with the number of shares you purchase (you can overcome flat commissions by trading more shares, but you cannot do the same for the spread which is obviously per-share.)
Any recommendation to learning the Financial side? I have tons of experience with data science algorithms (specialization) and computer science (major), but have little knowledge toward terms used.
Thanks for the comment. Like programming tutorials, there are many great (free) resources on trading/finance available online. Investopedia has some great content to get an overview, otherwise, you could check out my site TradeOptionsWithMe.com for some stuff on options (and equities). In the beginning, I’d just recommend staying away from overhyped trading guru products that promise to reveal incredible strategies. Instead, try to build an all-round knowledge foundation on the markets, asset classes, finance etc.
I made my own trading robot and tested it on old data and found that it was 80% effective, but when I tried it on real data, I noticed that it was not working, I don’t know why, even though I use a good server with a fast speed
Thanks for the comment. Note that there is a huge difference between historical results and actual real-life results. These differences can be due to biases, errors, market changes and much more. I have two videos going over some of the potential problems when it comes to analyzing backtest results that you can check out here: ua-cam.com/video/VmD2fUt8KYY/v-deo.html ua-cam.com/video/oFvYbfDOJ5c/v-deo.html (this one is part of my free algo trading course) I recommend using QuantConnect since they allow you to create, backtest, and host your algos in one place: www.quantconnect.com/?ref=towm In my course, you also learn how to create algorithms using Python and their platform.
QuantConnect since they allow you to create, backtest, and host your algos in one place: www.quantconnect.com/?ref=towm I have a course in which you also learn how to create algorithms using Python and their platform.
The problem is that when developing an algorithm, you usually don’t backtest it once and then you’re done. More often than not, you realize some flaw or possibility of improvement after backtesting. If you then use 8 years of the data for backtesting and developing, you’ll likely overfit your algorithm to that time frame. While developing, I’d recommend using a much smaller portion of the available data. Only, when you are completely done, should you start touching the rest.
I recommend learning how to code yourself. I know it can be intimidating at first, but it really isn't that hard. Not relying on someone else gives you so much more freedom and in general, programming is a skill that will always be useful (in areas besides trading as well).
I am a self taught programmer and have been so for 8 years. Its incredibly fun, good for your mind since its all logic, and very rewarding when something works. I really think you should teach yourself to code. Python is a good starter language and is great for data manipulation. Places like QuantConnect allow you to test the strategies you code and you can learn to code for free.
Except for TODO or FIXME comments, or API documentation, I haven't commented a single line of my code since I began with TDD back in -02. The tests are the documentation. If there is a piece of code that is not obvious, just check the test for that unit of work to see what it's supposed to do.
Haha would love to know why you think this. Sure Python is not very fast, but in terms of the community, libraries, data-science, and algorithmic trading support, I find Python one of the best languages.
@@TradeOptionsWithMe No type system, bad testing framework, 70x slower than c/c++/rust, horrible multithreading, no pre-compilation which leads to errors being found in the middle of a backtest after hours just because you passed in a wrong parameter type. I switched to Rust but had to write > 20k LOC from scratch so there's that xD
If you want to learn algorithmic trading, check out this free course that I created here on UA-cam: ua-cam.com/play/PLtqRgJ_TIq8Y6YG8G-ETIFW_36mvxMLad.html
What helped me tremendously was creating a playback engine in Python that uses recorded data to perform a "walkthrough", but without the restrictions of real time. So I record 1 sec data, and then play it back by reproducing the data flow as though it was real time. With this method, you can replay a full day as fast as your computer can rip through it. For me it is about 1-3mins depending.
Thanks for sharing!
do you think you can help me create my option spreads strategy
Thanks for the comment. I currently don’t create algorithms for anyone else. But if you want to learn how to code your own options trading bot, you could check out this video in which I code an example options trading bot: ua-cam.com/video/Lq-Ri7YU5fU/v-deo.html
Former professional tester here. I was writing up the same comment. The "backtesting" talked about in this video seems rather badly broken - or possibly it is intentionally crippled in order to save CPU cycles in the cloud instance.
I'm happy I developed a trading strategy for what I want my algo to do before I even started learning. Learned alot from this. I just started my algo trading journey and appreciate all your videos
Awesome! Thanks for the feedback.
how is it going?
Walk Forward testing is a technique where you backtest with staggered periods of optimised and out of sample data. I think what the OP meant to say is Forward Testing.
Not trusting your algo -- that's the biggest one for me, mainly due to psychological reasons. You need to make algos that fit your psychology.
Such valuable and honest content. Keep up the good work, man! Quick question, what video software do you use? Is it doodly? Regards.
Thanks! The software is called Videoscribe.
This is not what walkforward testing is! This is live (paper) trading testing. Walkforward testing is backtesting with a sliding window.
Other mistakes to avoid:
- relying on a single strategy vs a portfolio of not-too-correlated strategies
- not monitoring the live strategies
Hi Jul,
Thank you very much for your comment and mentioning other common mistakes. Paper trading with live data is a special case of walkforward testing. I should probably have specified that more clearly. Thanks for pointing that out.
Merciiii for sharing this! Really valuable
I would disagree that walkforward testing is a must. Instead I start out small but with real money. Every loss hurts and you analyze the trades much more. It also motivates you to continue when you earn real money.
Hello. This was a very good expose on the pitfalls of algorithmic trading. Here's my question: what if one's broker doesn't charge any commission and/or trading costs? What are costs that are still left to consider? Thank you in advance
Thanks for the comment. Often there are clearing fees that aren't mentioned by the brokers since they technically aren't broker commissions. But even with no clearing or broker fees, there are other costs. Slippage is one of the biggest non-obvious costs. Slippage is the money you indirectly pay by getting a worse fill price than your initial order price.
Otherwise, there also are market impact costs. Depending on your trade size, and the liquidity, your orders can directly move the price of the security. This market impact will usually lead to worse entry prices.
Depending on what you trade, you might also sometimes not be able to get filled at all. If, for instance, a security is hard-to-borrow, you might not be able to locate any shares to short. Or you might have to pay high margin rates/borrow fees.
I hope this helps. If you have any follow-up questions, definitely let me know.
@@TradeOptionsWithMe This seems to assume that the trader is consistently working with even lots and several lots at one time. But what about smaller trade sizes? I'm thinking maybe 100 shares or 200 shares. And does 'slippage' occur with a market order or a limit order?
@@William_sJazzLoft It really depends on what you are trading. If you are trading highly liquid, high-float equities with a few hundred shares, your market impact costs likely are quite negligible. But if you are trading iliquid, low-float penny stocks even a few hundred shares can have an impact.
Negative slippage can usually be prevented by using limit orders, but this comes with the risk of not getting filled which you can think of as an opportunity cost.
@@TradeOptionsWithMe OK. Although I don't generally deal with penny stocks, I do for the moment generally trade in stocks that are priced under $10. I have made remarkable progress having done so thus far. I'm on track to clear a cumulative profit of 10% in this next trading week. But I use limit orders almost exclusively ( both buying and selling ). I may have to be a little bit more patient - but I always get filled. And that was one reason I asked you the question about what cost might still be incurred. But still I appreciate the admonitions and advise that you are giving
Slippage and the spread are the largest trading costs that I have encountered. The video author has already covered the topic of slippage quite well.
The spread can be quite large in certain cases; for example, going long with the ETF TQQQ reveals fairly small spreads due to good liquidity, but SQQQ has much larger spreads because it trades to the closest penny but the stock was only about $5 a share at one point. A recent 5 for 1 split helped this problem, but after a market rally the spread problem is starting to appear again (and cut into profits). Your algorithm has to, on the average, make more profit per trade than the spread. This is easier said than done, especially for day trading algorithms that trade frequently. I found that the spread was a much greater obstacle than trading commissions, since it scales with the number of shares you purchase (you can overcome flat commissions by trading more shares, but you cannot do the same for the spread which is obviously per-share.)
One of the best videos on this topic! Cheers
I enjoyed your honest truths about Algorithmic Trading. Good work.
Thank you! I am very glad you enjoyed it.
Thank You for the video
very good explanation, thanks a lot for your job
That's quite useful! Thanks, mate!
Thanks for the great points!
Thanks Mike
finished watching
Thank you!!!
Any recommendation to learning the Financial side? I have tons of experience with data science algorithms (specialization) and computer science (major), but have little knowledge toward terms used.
Thanks for the comment. Like programming tutorials, there are many great (free) resources on trading/finance available online. Investopedia has some great content to get an overview, otherwise, you could check out my site TradeOptionsWithMe.com for some stuff on options (and equities). In the beginning, I’d just recommend staying away from overhyped trading guru products that promise to reveal incredible strategies. Instead, try to build an all-round knowledge foundation on the markets, asset classes, finance etc.
Very well explained - thank you
Glad you liked it!
Very useful!! Thanks for the video
Very good work! Thx
Excellent!!!
Thank You!
good good good. thank you sir
Really great. thanks.
Yep I agree, Demo accounts have lower spread than real accounts.. Plus some live accounts have floating spreads👆🏽
I creat a robot on fxdreema can I back test it on quantconnecct?
Nice video.
I made my own trading robot and tested it on old data and found that it was 80% effective, but when I tried it on real data, I noticed that it was not working, I don’t know why, even though I use a good server with a fast speed
Thanks for the comment. Note that there is a huge difference between historical results and actual real-life results. These differences can be due to biases, errors, market changes and much more. I have two videos going over some of the potential problems when it comes to analyzing backtest results that you can check out here:
ua-cam.com/video/VmD2fUt8KYY/v-deo.html
ua-cam.com/video/oFvYbfDOJ5c/v-deo.html (this one is part of my free algo trading course)
I recommend using QuantConnect since they allow you to create, backtest, and host your algos in one place: www.quantconnect.com/?ref=towm
In my course, you also learn how to create algorithms using Python and their platform.
Can you tell me what is the ideal server to try algo
QuantConnect since they allow you to create, backtest, and host your algos in one place: www.quantconnect.com/?ref=towm
I have a course in which you also learn how to create algorithms using Python and their platform.
Can you give us any EA out there that work well we can use?
No I can’t. I recommend trying to build your own automated trading strategies.
what if we have 10 years of data and we backtest on 8 years and forward test on the remaining two?
The problem is that when developing an algorithm, you usually don’t backtest it once and then you’re done. More often than not, you realize some flaw or possibility of improvement after backtesting. If you then use 8 years of the data for backtesting and developing, you’ll likely overfit your algorithm to that time frame. While developing, I’d recommend using a much smaller portion of the available data. Only, when you are completely done, should you start touching the rest.
Does anybody know where to find a programmer to create the algorithm? I dont know how to code
I recommend learning how to code yourself. I know it can be intimidating at first, but it really isn't that hard. Not relying on someone else gives you so much more freedom and in general, programming is a skill that will always be useful (in areas besides trading as well).
I am a self taught programmer and have been so for 8 years. Its incredibly fun, good for your mind since its all logic, and very rewarding when something works. I really think you should teach yourself to code. Python is a good starter language and is great for data manipulation. Places like QuantConnect allow you to test the strategies you code and you can learn to code for free.
Couldn’t agree more!
@@abethepleb7915 thanks alot bro
Except for TODO or FIXME comments, or API documentation, I haven't commented a single line of my code since I began with TDD back in -02. The tests are the documentation. If there is a piece of code that is not obvious, just check the test for that unit of work to see what it's supposed to do.
Biggest algo trading mistake: Using Python!!! I might get hate for this opinion but sooner or later you will know what I mean.
Haha would love to know why you think this. Sure Python is not very fast, but in terms of the community, libraries, data-science, and algorithmic trading support, I find Python one of the best languages.
@@TradeOptionsWithMe No type system, bad testing framework, 70x slower than c/c++/rust, horrible multithreading, no pre-compilation which leads to errors being found in the middle of a backtest after hours just because you passed in a wrong parameter type. I switched to Rust but had to write > 20k LOC from scratch so there's that xD
Those are certainly some valid points. But especially for beginners, writing 20k LOC isn’t really viable xD
Thanks Captain obvious =)
my algorithm can't even make profit in training data 🤣