Ask HN: Before launch – everybody says "dont connect the trading engine" – why?

[ta12653421]

Hi HN Community,

during the last couple of weeks i spend a lot of time on building a fully automated trading engine. Some kind of "disclaimer": im now in the financial space since ~13 years, working for startups, creating new banks and working for them.

While i'm not an "equity guy", i thought there should be some potential when playing around with some of the data feeds: it should be possible to build a fully automated bot, which is just a simple trend-following model and the system should just jump on running trends on whatever mega-corp stock, go with them some days and then exiting.

So i started to build an engine which contains all the things i learnt during this interesting journey: the system reacts dynamicly depending on different market structures, it also integrates events (like earnings dates etc), it goes long and short, it captures trends for only 2 - 3 days sometimes.

Now, im that convinced that this application should work under real market conditions; during development, i also made sure that the system works under "live conditions", meaning that data is added incrementally to the algos etc. (i found out there are a lot of wrong calculations/indicators/chartings out there which are failing if you connect them to a "real" datastream, since very often a massive mistake is done during development: presenting the "data as a whole", which leads to perfect results in testing, but fails completely under live conditions etc.)

A special perk with the application is the following: in my country, there are special leveraged products available (turbo warrants), which are just replicating the increase/decrease of an underlying by given scale/gearing, just that simple. Those are nearly risk free if you hold them only for a couple of days with a decent lever/gearing. I say nearly risk free, since, for sure, the position may crash if a completely unknown Unknown appears (like a military event or similar) Applying a leverage to the trading behaviour magnifies the results massivly, it should easily be possible to get 400-500% per year. The math is very simple: the system is shooting at currently 150 symbols across NASDAQ100 and SP500, jumping onto the short-term trends, regardless if long or short, and applying a lever/gearing of 3 - 7 and then reinvesting, rinse and repeat for one year.

Why automation is the only way to win as "active market participant": Most people have 3 - 10 stocks in mind when thinking about investing, and maybe they are following 20-30 items on their watchlist "which they watch from time to time" - if they are watching it at all. i have a different perspective: analyzing the biggest stocks worldwide on a daily base an computing my entries/exits, there are opportunities every day. Its not about "doing this one big trade", its about continuous and reliable but smaller returns on a reliable basis: The system is handling around active 30 - 50 positions under normal conditions (and my starting budget is limited...) Automation on a larger list of strong stocks is the only chance to outperform, in my oppinion.

Now i showed all of my ideas, results, data and code to some people to get some feedback:

- all of them are "equity guys" and working in the field as "investment fund manager" or similar or they are working at larger banks in whatever departments

- none of them is a CS-guy understanding anything of coding etc.

ALL of them say that i shouldnt connect my system to the real market since it will quickly loose money - and the only answer, when i dug deeper, was: "...because there have been others before you who thought they were smarter than the rest..."

So - who is wrong? Is it me and i oversaw something critical?

[charleslmunger]

>I say nearly risk free, since, for sure, the position may crash if a completely unknown Unknown appears

I would encourage you to read the books "Antifragile" and "The Black Swan" by Nassim Taleb. The risk you're running may have low probability, but also has massive effects when it does occur, and such events are responsible for a major part of both gains and losses in financial markets.

>"because there have been others before you who thought they were smarter than the rest..."

Short term trading without a latency advantage is a zero sum or negative sum game. Every opportunity your bot considers is also being considered by competing bots, with way more capital than you. Maybe you are smarter than the other market participants, but it sounds like maybe you haven't done your research on how sophisticated other bots are. It's a bit like betting on yourself in a sports competition before finding out how your skills compare to the people who regularly win those competitions. However, unlike sports where you'll be immediately trounced by a professional, you might make some good returns on small trades to start, which will make you feel comfortable putting your capital at risk, and then you'll lose it all.

[ta12653421 → charleslmunger]

Hi, thanks for your feedback!

> I would encourage you to read the books "Antifragile" and "The Black Swan" by Nassim Taleb. The risk you're running may have low probability, but also has massive effects when it does occur, and such events are responsible for a major part of both gains and losses in financial markets.

I understand where you are pointing out at: First, the engine exits on distinct loss-thresholds, reducing the loss. Sice stocks mainly move in (sometimes only short-term) trends, i do not see why such an instrument should include more risk than an the stock itself, apart from the fact that the loss may be multiplied. With "risk free" (hence the quotes) i'm referring to the risk of complete loss/total crash of a position. For sure, the system does not invest all-in into one symbol! :) (max 2% of the total capital, according to std. risk management practices)

For sure: None of the systems will be always right, but if your quote is good enough, then there is a high chance of making very good money; the system is not about "predicting the stock price" on some fancy ML stuff, its just a solid trend-following algotrithm. And yes, the system will loose money! As every "manual" trader does. The advantage is mainly that such a system is capable of managing a lot of positions and tracking them - you could do this on your own, manually, but its exhausting/boring/prone to errors etc. And yes, there are a lot of unpredictable events, therefore risk management is very important and at the core of the system.

I like the comparison to sport betting - though i think its not correct: In gambling, you either win or loose - but on the stock market i can exit and minimize my loss early enough to not to loose everything. (i have never done a sport bet, nor will i do - but this is my understanding of sports betting? its like "binary options" in financial terms, an instrument i would never touch...)

And: large mega-corp stocks are not that volatile to be in a higher risk than buying the stock itself: a big-co stock has a volatility of ~1 - 2% per week (ye, Nvidia is an exception!), this volatility is absolutely in a range i can sleep better with.

Another comparison: What is more realistic? Scenario A: you are getting fired tomorrow by your boss since he doesnt like your fresh haircut? Scenario B: that 50 - 60 stock market positions on the 100 biggest companies globally are crashing all at the same day?

Actually, i can sleep much better with the latter dependency than with the former one.

> maybe you haven't done your research on how sophisticated other bots are. Maybe not as solid as you did, but my question is: Where are all the highly profitable quant bot owners if there are so super-alien-sophisticated strategies which much more "intelligence"?

After doing my initial research, i come to the following conclusion: A) mighty banks & MFI have better knowledge/knowhow/infrastructure and they are the only ones who are moving the market B) most people claim to have a super-mega-system while they are actually cooking with water as everybody else C) simpler systems works much better than any voodoo-magic-boohoo implementation: a friend of me also did some ML-implementation these days on the same topic with some latest research papers: his numbers are not better than mine, while applying a muuuuccchhh more complex tech stack

And my last 2 cent to the sports betting comparison (since i have heard it quite some times): I'd say most games are manipualted, esp. in smaller leagues - while it is unlikely that someone is big enough to "manipulate" NVDA or QCOM, id guess?

[charleslmunger → ta12653421]

My comparison on sport betting was that you're betting on your own performance, which you know a lot about, but your opponent's performance is just as important. If your downside is really capped, not just in terms of a stop loss which is really an automatic sell, but actually capped as far as your liability with the type of warrants you're using, then by all means run your bot. I suspect you're far enough invested in this idea that you'll always wonder if it would have worked if you didn't try it. Just don't become the next LTCM.

[ta12653421 → charleslmunger]

> if your downside is really capped, not just in terms of a stop loss which is really an automatic sell, but actually capped as far as your liability with the type of warrants you're using

good catch! actually, its this category of products, they are mainly common in AT, DE and CH: https://en.wikipedia.org/wiki/Turbo_warrant

this category is actually a "retail gambler product", as i would call it - in most countries, they are heavily regulated by the fin authority, though they are mainly used by private persons - and even the regulators in most countries says that retail customers mainly loose money with them because they are applied by uninformed investors etc. and compared to >classic options<, there is no margin call or similar or comparable hooks - investors can just loose what they put on the table, nothing more.

(if you may ask: no, these are not meant for long term investments as the price decays over longer time, this is how the issuer is financing this instrument - in fact, these instruments are exactly what i need to run my strategy: the decay happens once a month by a small percentage, which is known upfront - but using them for 5 - 15 days is fine and comes more or less at no cost, as long as the position is not going against you)

Important: This is not comparable to CFD - which i consider a real "gambling product", since its absolutely intransparent and you are victim to the issuer, e.g. under certain market conditions you cant sell the position somewhere else e.g. (CFD are that a robster-product that platforms for trading EU in EU them have to put a risk-disclaimer on the website and the text for this disclaimer is defined by the regulator :-D)

> you'll always wonder if it would have worked if you didn't try it

ha! THIS is exactly why i love HN, this is my throwaway account: its actually that thinking of the community over here! :) :) thats also my perspective: I will never find out if i dont do it! So relly, yes....

> Just don't become the next LTCM

Unfortunately, this strategy has its specific limits - e.g. you cant scale it unlimited, apart from position management and market chances: things like deposit insurance etc., you would need to split onto several acconuts (which may not hold in case of bankruptcy at the same institute), number of brokers is limited (currently, the app works only with one specific broker/bank, as they are the only one with a full REST interface for their trading desk)

E.g. there are chances every day, but there are not unlimited chances every day - e.g. what i can see from the testdata is that the bot quickly runs out of potential trades after a couple of month, max open positions (with money left in the account) is around 50-80 with NASDAQ100 + some SP500 titles.

So it should end up as small cash mashine but you cant withdraw a million each year :)