Why building profitable trading bot is hard?
I wants know why 90% of trading bots are fails. What's the reason behind it.
Lol! Because you only hear about the bad bots.
Think about this --- if you had a really good bot capable of actually being consistently profitable, why would you ever advertise it or tell anyone about it or even acknowledge it's existence?
What I found out that my trading firm is a market maker, and this trading firm, alongside with other non individual (corporate) players in trading world, have moats in their cutting of fees.
To make a profitable trade, you need to take into account many of the fees (and taxes) or else you get into death by thousand cuts. Trading firm simply operates in a different fee structure than individual traders. Some market makers are actually getting paid instead, just by providing liquidity. That’s why they can be profitable.
If an individual stock trader buys stock A for $100 and sell it later for $100, it is considered loss (due to fees) or at least break even. Not so much for trading firms, they actually make profit.
Since trading bots are just individual traders, same rules apply.
CMIIW
Also, yes market makers can get left with bad positions if the market moves against them, but when I was in the industry operational failures were the much more troubling issues.
All of its trading though and none of this is new.
With the above, you'd think that a trading bot would be a piece of cake. The easiest part of the process, probably. As of right now, I've found that it's hard to model intuition using code. It's easy to watch a graph and say "I think it's about time to sell", but it's hard to give that understanding to a computer. I do believe that it's a problem that can be solved, and certainly you can look at Aladdin[1] as a system that has been very successful in this space.
Don't take negativity as a reason to not do something though. If you're passionate and interested, then go prove everyone wrong.
[0] "per day" references per day that Feetr finds a stock
Any suggestions for doing this?
If you can't do this; if you sell too early or too late, your return will potentially be a lot less than picking the high price after the fact.
In other words, the stated returns are maximized assuming perfect clairvoyance --- which no one has. Basing return on closing price assumes zero clairvoyance --- which the average person has --- and is thus a more realistic measure.
The metrics are as they are because, right now[0], there are no better options to validate how Feetr performed. These values are, like you say, theoretical maximums and have not been matched by humans in almost a year. You get the top every so often but not consistently.
However, I wouldn't say that this makes the service less useful than people think. I have made 700% returns and this allowed me to quit my job and work on Feetr full time. Other beta testers have made more than their salary using this.
[0] As above, there will be an investment service offered at some point
All data is 100% available on feetr.io after market close and, when it launches, Monday will be given for free on Reddit, Twitter, and to anyone who has a free tier account at feetr. Monthly subscription will be $8.99.
I would offer it for free but I believe that it's immoral (and potentially illegal) for me to be able to invest in the stocks when it launches, and so I've priced it at a point that I think people can justify. I tried to match Netflix on price, but they increased it so now I can say I'm cheaper than them.
If investing on Monday is making people more than $9 then it's worth it to subscribe, and if it isn't then I need to rethink the algorithm.
Great marketing here, the tool is so good that it would be "immoral/illegal" for the creator to use it.
But when it launches, I have a competitive advantage over everyone using the service and that doesn't sit right with me.
It's easy to pick the high/sell price for the day after the day is over. Predicting this on the fly is a whole lot trickier. It's very unlikely you will catch the day's high. Unless you're psychic, you'll probably sell either too early or too late with diminished returns..
The concept that I try to push is to focus on compound interest. To start with a small amount and just keep rolling it over into (near) daily profits. Don't try to time the top, sell when you're comfortable.
And I appreciate that a lot, however I prefer monthly billing to ensure that subscribers are never locked into the service.
It’s my personal opinion, based on my time and experience (and several books about the field), “90% fail” in the same way “90% of business fail.”
Tons of reasons but I truly believe lack of alpha is overestimated. I think finding alpha is not nearly as hard as people make it out to be. Execution and continuously innovating is the real challenge. Design your systems, knowing you will have to change up your strategy and focus on a system that will help you find new alpha.
I believe after lots of research, this is how Rentech did it.
And to touch on the business ananology again, you could have the best donuts in town but if the rest of your business is fundamentally flawed, it will become part of the “90%.”
Now, I’ll say this, I don’t manage some insane portfolio or anything. However, I run a private firm (myself and family) that makes for a comfortable life. I work more than 9-5, but I enjoy the challenges and freedom.
If you’re asking about more complex bots used in managing larger, institutional funds, I’m afraid I can’t answer that… but I imagine the challenges are quite different.
I try to post things that have helped me with building these systems from time to time.
Lmk if there’s anything else, I can try my best to answer.
I actually don’t know fundamentals well at all and rarely keep up with news.
The systems I designed are for trades lasting no more than a few hours, so much of that stuff never comes into play.
I’ll just say, I think investors like Warren Buffett are good to emulate. They play the long game and play it well… I think doing something like that is a full time job though.
Last note, just knowing a few ppl personally, I’ve seen people grow their wealth by just sticking it into the market and forgetting about it ;)
Just look at compound results of hedge funds (published by Goldman Sachs). These smart quants are consitently buying high and selling low for last couple of months and years.
[] GSTHHVIP vs S&P500
Therefore, a trading bot would need to trade on information that is not available to other bots, for it to be profitable. This is hard to achieve, and each bot that _do_ achieve it is making the market even more efficient.
ETFs (Electronically Traded Funds) have no fund management fees; like Class B stock, ETFs typically are not voting shares (which you don't have when you buy a mutual fund anyways).
Algotraders reference e.g. an S&P 500 ETF as the default benchmark for comparing a portfolio's performance.
Quarterly earnings reports are filed as XBRL XML. A value investor might argue that you don't need to rebalance a portfolio until there is new data about technical fundamentals for technical analysis.
The average bear trades on sentiment and comparatively doesn't at all appropriately hedge; this is part of Behavioral economics, the technical reason why some people actually can outperform the market, imho.
Here's the awesome-quant link directory: https://github.com/wilsonfreitas/awesome-quant
So any regularly profitable trade of any substantial volume would be copied pretty quickly, squeezing out any profit.
No doubt one could find an obscure trade, keep volumes low and stay under the radar long enough to make a tidy sum.
Because you have to find a pattern to build it on, and patterns change frequently. If you're wrong or it changes, then you lose a bunch of money.