A machine learning approach to stock trading | Richard Craib and Lex Fridman
ฝัง
- เผยแพร่เมื่อ 8 ก.พ. 2021
- Lex Fridman Podcast full episode: • Richard Craib: WallStr...
Please support this podcast by checking out our sponsors:
- Audible: audible.com/lex to get $9.95 a month for 6 months
- Tryolabs: tryolabs.com/lex
- Blinkist: blinkist.com/lex and use code LEX to get 25% off premium
- Athletic Greens: athleticgreens.com/lex and use code LEX to get 1 month of fish oil
GUEST BIO:
Richard Craib is the founder of Numerai, a crowd-sourced, AI-run stock trading system.
PODCAST INFO:
Podcast website: lexfridman.com/podcast
Apple Podcasts: apple.co/2lwqZIr
Spotify: spoti.fi/2nEwCF8
RSS: lexfridman.com/feed/podcast/
Full episodes playlist: • Lex Fridman Podcast
Clips playlist: • Lex Fridman Podcast Clips
SOCIAL:
- Twitter: / lexfridman
- LinkedIn: / lexfridman
- Facebook: / lexfridmanpage
- Instagram: / lexfridman
- Medium: / lexfridman
- Reddit: / lexfridman
- Support on Patreon: / lexfridman - วิทยาศาสตร์และเทคโนโลยี
![[TH]FFWS SEA Spring 2024 Final](http://i.ytimg.com/vi/TK2mu-8yBFU/mqdefault.jpg)
There are no trading strategies (including data) that work that you can just buy, for two reasons:
- Nobody sells a working tradong strategy because a working trading strategy is worth more than money you can sell it for.
- Any advantage that a working trading strategy has will quickly disappear as soon as it becomes known to others.
Surely this applies less to trading algorithms which are constantly reanalysing data and results, and changing overtime to fit/further understand market conditions and other variables that would normally erode any given edge overtime.
"Stay high for as long as possible"
Great work brother Lex!!!!!
great content and insight
Good thing Richard Craib was coherent.
Does Numerai work with currency market?
Interesting idea but having an intimate understanding of the data is the most important pre-requisite to attempting any type of ML approach. Not being able to read the column names or contents is a massive handicap. Removing human intuition from feature engineering is like asking a doctor to cure a patient without seeing them.
It reminds me on the Goldcorp Story and Don Tapscott many years ago.
Where can i get this
I need help with the Finance market so bad since im a beginner
I like the idea, but don't see it working in the long term because of game theory principles. Personally, I see more value in preserving and executing my own alpha rather than submitting signals to numerai. However, I will still run my trading system on $nmr because regardless of whether I'm right or wrong, profit is profit.
I agree I wouldn't want to give my strategy to numerai either
@@pineapple3832 you're not giving your strategy to numerai, you're uploading predictions. They have no idea how your model is working and what the strategy behind these predictions is. That's how I understand it at least.
@@mahdif1000if they don’t, how is it that they verify the uniqueness of your model?
I wonder how they guarantee that poor performance can be attributed to the model as it might also be due to a large difference in data given and live data. On top of that, ensembling all these models does not seem arbitrary. Interesting stuff though, will look into this more!
How is a machine learning approach any different from a 'traditional' time-series forecasting approach to stock trading?
In many ways they are similar because they can be reduced to a form of statistics.
In this context I think a good traditional example is using moving averages to approximate future prices. A simple moving average has two features, time and price. Also the function of the formula for SMA is of fixed value.
A machine learning algorithm could have practically infinite features. Those features can be things price, time, volume, the sentiment of the news, and quantity of hashtags on a ticker. Additionally a machine learning algorithm could continually rewrite the function it uses to predict prices.
Successful quants have other aspects added to machine learning: signal processing and data fusion at least. These are topics in electronic engineering used in radar tracking, audio and video processing. Another aspect is decorrelation, which is not an easy topic. I remember Jim Simons in one interview saying they use neural network without feedback, a very new and not common approach, i guess because they dont want to feed back the errors inside the learning network again in the network. I guess they prefer to reconfigure the network in real time with continuos flows of data.
@@dfolegna that is interesting about the NN without backwards propagation , but there was a paper released by Hinton forward-forward but almost adversarial in that one forward pass learns one objective and the other is a discriminate but not two distinct networks like GANs. I wonder how Jim’s team learn the model/weights without the feedback
if it's over 3 months this doesn't make sense. events like covid and other crisis greatly influence it (and you can't predict that with stock data)
Truly brilliant. Been trading for 32+ years. And this is. Spot. On.
Lair
Hard to believe his “data” is anymore special than data found on the web. Anything that is publicly traded is basically public info. Furthermore, stock price only has a few data points per day: open, close, high, low, volume (adjusted close if needed) making it a minimalistic dataframe. In the end, if people attempt to “solve” the data, they’ll end up adding columns like moving averages, stochastics, OBV, RSI, MACD, DMI... ending up backtesting reversal/pivot points based on those metrics interacting with one another. If he adds market beta like jobs payrolls, FX flows, US treasury yield curve, he’ll end up with 10s of linear regression models (not even considering polynomial lines) that attempt to give a signal at the same time. Seems like he is trying to reinvent the wheel when the wheel’s already been invented. Good luck nonetheless.
@@msa7020 Sorry to disappoint but only scam bots will reveal to the world how much they earn daily and talk to themselves on how grateful they are to Linda.
All the "zero fee" trading platforms basicly make money by selling trading data of their clients to "high frequecy" trading bots ownd by big players who will trade against their users and have cheap approach to market.
@@fpxy00 I don’t recall him mentioning client data at any point in time throughout the interview. But if he does imply it then you’re spot on and this is in no way ‘solving’ the market, it’s just hedging away customer trades for the house to win.
Ah you have no idea. There’s so much to learn from the data you just named. So many ways to use the open high low and close algorithmicly. Man 😂
@@noobbeatswhat are the many ways to use them? I would be curious to know
How would you predict that Apple would get back Steve Jobs?
Technical and fundamental analysis is the foundation of profitable trading. Price and volume show where big money flows. Unless your managing huge amounts of money, you can jump on for a ride where the huge institutional flow is heading
Data is important because it allows you to predict the future with higher accuracy and steer your operations towards that direction and capture unmarked territory and profit from it first. Like Christopher Columbus. And now with the recent developments of Machine Learning Ai models and new optical hybrid CPUs and GPUs we are getting closer to an intelligence faster and possibly greater than our own and that's not taking into account NeuChip Biological Ai and Neuromorphic Computing.
People no longer visiting companies to see and understand their daily function, how they create value? If so, it's only speculation which is vulnerable to manipulation.
Ben Goertzel‘a SDAO will bring ai trading to everyone as of December 2021
Lex misspoke "reduce exposure to red lights" because he probably just spoke to Andrew Huberman about red light therapy.
Basicslly Momentum exploited onto Other Features?
Computer algorithm computer learning 101 AI has been doing this for year's
I am wondering if they have data also for the UK market. Does anyone know?
This is from their FAQ "Richard Craib has stated that the training data represents global stock market data."
You the man lex
They're 'giving away data'.. That's just a marketing way to put it
I wonder what his or Numerais opinion on the 10 year yield is!! 😱🤑
Binary code
'Amazing video, A friend of mine referred me to a financial adviser sometime ago and we got talking about investment and money. I started investing with $120k and in the first 2 months , my portfolio was reading $274,800. Crazy right!, I decided to reinvest my profit and gets more interesting. For over a year we have been working together making consistent profit just bought my second home 2 weeks ago and care for my family".
Hi. I’ve been forced to find additional sources of income as I got retrenched. I barely have time to continue trading and watch my investments since I had my second daughter. Do you think I should take a break for a while from the market and focus on other things or return whenever I have free time or is it a continuous process? Thanks.
@@theresagarcia1218 Quitting may not be the best approach if you ask me. This is where an AI comes into the picture. I barely have time to trade myself as my job swallows up most of my time. 'MARGARET ANN WARNKEN", a licensed fiduciary whom has made me over 5 figures in profit in less than seven months, handles my investments. I could leave you a lead if you need help.
@@mayacho4910 Oh please I’d love that. Thanks!
@@theresagarcia1218'MARGARET ANN WARNKEN"
@@theresagarcia1218Lookup with her name on the webpage, omit the quotations though.
I remember when lex used to breakdown the bjj for yoga guy’s fights.
For people that want to go down trading rabbit hole, don’t it’s a wast of time.
You need to read the reports and listen to earnings calls.
AI can’t preform high level judgment, the only way robots might make you money is trough arbitrage.
All models are wrong some are useful.
You are correct in that all models are wrong, only some are useful. The human mind and the best models we have to analyse stocks or nature etc. are a great example! The same goes for virtually everything one can concieve. Humans will always be there to perform high level judgement and improve these models, but as AI becomes more advanced and as we improve the data it uses, that judgement will be used largely to improve the techniques and scope of said AI, rather than manually tweaking the underlying models ourselves. An AI that is continuously processing and reprocessing an increasingly large number of data sets (price, indicators, sentiment, fundamentals, even its own trading results and more), training itself on new variables and effectively discovering trading systems on its own with constant human improvement could very well create a consistently successful trading system (until an unforseen variable not yet calculated comes into play, which it could then incorporate).
To a degree, what I described above is essentially what professional stock traders already do! They just traditionally use their brain (a neural network trained on trading systems and life experience) and analysis software to speed it up. Any edge that a human trader has over an AI trading system is largely due to the data sets they have available to them (fundamental analysis, or social factors not yet fully processed/understood by machine learning) and the superior computational power of the human brain.
Algorithmic trading is never going to be set and forget. Constant iteration, supervision and expertise is required to improve the accuracy of the AI as it slowly understands the sheer number of factors (human and technical) that are involved with the markets.
@ 9:40 what he means is the more you expose your Model to obstacles and set backs and failures the more it will learn adapt and develop. Similar to humans
due to EMH and you do not have any sector knowledge you are way better to trade cross sector. If you are a medical phd then fine stick to health sector...
I'm not sure about this bro...And yes I trade HFT( with video proof )..Not hating, but this is not how HFT works. The root data come from the exchanges. The first access does cost but not millions. exchanges charges for the first right of distribution. Then next layer pays for access. Normally this access is categorized into 2 distinct groups ... retail and professional. There is all sorts of data, but if you think interference of data (obfuscated or not ..LOL) will help you see an iceberg order or tell you when to jump to an inverted exchange to capture a rebate...I'm not sure about that. The order types at the professional level offer more options than you can imagine. For example, I can place an IOC order at a "WHEELed" algo that will automatically cycle every exchange until the entire order is filled via rebate. I could go to an exchange that will automatically pull my order if I'm at the back of the order book ( yes certain exchanges offer this capability )...Will data see that? Regression, RNN will give you a high probability but no guarantee. The best way to approach HFT is to concept winning scenarios (non-obvious) and mix and match the right exchanges, fee structures, and even book-leaning. I had better stop now, But just be careful about folks claiming to have the magic bullet...Hint Place orders early ( up to 1 day early, rest eval position, and pull if needed) Make your own!
Yeah, the 3 month holding time HFT strategy of Numerai.
Maybe watch the video before posting how it's wrong?
@@AnthonyBerlin Didn't say it was wrong ... don't really care... I was just stating how true HFT works.
@@GambrellTv Good for you I guess? I mean, noone asked about HFT and the video wasnt about HFT and you literally said "Not hating, but this is not how HFT works".
HIIII YAH!!!! All good@@AnthonyBerlin I wish you the best buy LOW, Sell HIGH ..no matter what!!! Have a good one!
@@GambrellTvaye you get it
Lex tha bro
Eminem just became really smart
Genius. Side note: Richard would look cooler bald. 👍
Historical data is useless for stock price
he's not referring to historic data, he's talking about features that affect a stock price, if you can implement many of the important features, you can get a legit idea if a stock is good or bad for a short time hold
This guy is smarter than me so I’ll just make fun of his hair.
love you
If this guy plays his cards right, he'll be rich af in no time
Their performance sucks, so I would say that's highly unlikely
Stay high as long as possible😂
Tying to learn how to trade automatically without learning how to trade manually.... brilliant... (sarcasm) You'd be better off to hire a bunch of trainees, pay them minimum wage and fade them.
;)
I literally can't tell if this guy has a british or american accent
sounds South African/Australian/New Zealand to me. Reminds me a bit of Elon's accent.
This approach will have limited success. First problem is only looking at 10-15 yrs worth of data which is essentially a long bull market. Other problem is with anonymizing the data is that you can't derive any meaning. Understanding what the data means is essential to making a decision and ML/AI is still to basic when it comes to understanding. Understanding the data is how macro economists (Dalio, Drunkenmiller) and subject matter experts (Buffet, Chamath) get their edge.
Drukemiller, drunkenmiller, lol.
denial hair
💀
🤣🤣🤣🤣🤣🤣
Investing is counted in years...not months...what a fool he is.
with the front of machine learning and algorithmic trading this guy is running a crypto scam
he might but it’s still possible. just keep testing until it works
He talks about 2% being a lot. Already there you know he is a scrub.
It's not investing - it's trading.
You create no value - you extract the value that others have created.
Focus on something that benefits humanity, not your bank balance.
You fool go focus on becoming a hairdresser to help humanity. With this knowledge you could build schools for children instead of being one teacher, you could pay 100 teachers for years in low paying countries.
Your mindset looks like the mindset of a lazy fake Saint
Trading creates value by creating liquidity, ur also taking risk, and ur giving money to the companies at the end of the day
@@rafsanimam5647 Yes, you are making a profit for the companies, but that is NOT value. Risking your money to create a profit is also NOT creating value. No value is created at any point.
@@the-selfish-meme7585 i make profit then use it for consumption driving economic growth or i invest it into other companies so they have funding for innovation. Im still creating value