I may be way to cynical for my own good, but there's no proof that this is actually real, right? Besides a couple of pretty bizarre screenshots [1]. I'm skeptical because a) these types of gains in public markets are pretty much unheard of, and b) faking a story like this would be a fairly easy way of getting lots of attention for your freelancer business (which this guy advertises right at the top of the article).
[1] Running cat on a csv file would usually just print a list of numbers, yes? Instead he gets ascii art with months on the vertical axis, making it seem like time going back and forth, and some random commentary in the right margin. I dunno but something about it screams "mockup". As does the minimalist, ultra-stylish AWS window.
Edit: also, if it were me who got incredible returns from trading at high speed on indian markets using only machine learning on historical data, I would definitely keep quiet about it. Just sayin'...
There's no cat and no bag - just statistical sleight of hand. The people playing the trick know this, and are just collecting investor dollars while they can (and astroturfing every HN article that explains what it is). Relying on the output of chatgpt is no different than doing a rain dance and expecting it to rain.
To illustrate it, I've redrawn the green and red bars differently to show that, lo and behold, there are another 15 years of bull market in our future.
I love a good prank, but imagine a stock broker or forex pulling something like this. It probably looks a whole lot like market manipulation since it could lead holders to buy or sell based on a false trend.
It certainly makes me suspicious of the charts and claims forever thereafter if it is that easy to post fake info on a whim.
If you're not seeing it then I'm not going to argue with you, neither of us has the data to prove any of what might be going on.
However multiple trades of very similar amounts in under a second, after a long, massive slide in much larger amounts in the morning, looks very dodgy to me. Especially when these trades appear at the last possible moment, when all market resistance to freefall has just disappeared.
The experts in AI speculation over at Manifold Markets give this a 35% chance of being legit. There’s some additional theories in the comments over there.
The small subsets and super high Sharpe ratios look suspicious.
Further red flags in this particular case:
- Completely unclear what kind of data they're using. Are they assuming they can buy and sell one individual contract at the bid price each minute? Or did I miss some crucial information about bid-ask spreads?
- Abstract mentions a profit, not an information ratio/Sharpe ratio or anything similar.
- During training they need to tweak the reward function in order to not end up with "buy and hold"? How good is their strategy compared to buy and hold?
LOL? An individual claiming automated trading resulting in >100% annual gains is a fairly extraordinary claim. Triple digits in $ is a hell of a lot easier to believe.
The mention of 81 exchanges left me thinking: If none of the trading was fake, but then I created a single exchange literally named "ScamCo Faketrading Exchange" which claimed 1 petatrades per second were happening on its platform, then would this analysis say that >99% of all trades were fake?
I think it would.
I don't think any of the exchanges that I could have named off the top of my head were on their naughty list (paywall is keeping me from loading it again to check), ... so it seems to me that this report might be a bit overstated.
That paper uses simulated (i.e., fake) market data. The quote is just a bare assertion, not a conclusion. It's very difficult to identify spoofed orders in market data unless you're the exchange and know who sent all of the orders.
>Specifically, Nanex saw a spike in the milliseconds before 9:54:58 on December 7th, 2012. To be exact, they saw a flurry at 9:54:57.18, nearly a full second before the "third-tier" algorithmic subscribers got their data at 9:54:58 a.m.
This is exactly what you would expect to see if someone, or a bunch of someones, had access to the data even before 9:54:58 a.m. In this game you would want to hold your cards until the last possible moment before placing your bets.
A phenomenon not unlike what many of us have experienced bidding for an item on ebay.
Yet another example of the game being rigged. I don't invest for a living, but I've always thought it folly to approach the exchange in any manner other than a long term diversified one (as an individual investor). Maybe it is my lack of sophistication in the area, but anything else feels like gambling to me.
There's no need to believe, it's all public data and many thousands traders/bots are using those to trade every day... if the orders were fake it would be known and arbitraged very quickly.
Talking here about all the major exchanges with actual volume (Coinbase, FTX, Binance, etc) of course might not apply to some scammy third-tier exchanges with no volume.
I think this is bullshit. The facts on which it's based appear to be:
1. Someone working on high-frequency trading at Goldman Sachs was a Bad Person, as demonstrated by the fact that he seems to have leaked a big pile of GS's trading code.
2. The NYSE put out one report in which GS's trading figures were bizarrely omitted. (They've since blamed it on a "system error"; who knows whether they're telling the truth?)
From which the author somehow concludes:
3. GS probably hacked into the NYSE's machines or connections so as to be able to snoop on everyone else's trades before they committed.
Now, sure, #3 is possible. But the only evidence our author sees fit to provide is points 1 and 2 above, plus the observation that it sure would be profitable to do #3 if you could get away with it. (The reasoning, if you can call it that, seems to be: #2 suggests that there's more going on than meets the eye, and maybe GS were removed from the report because they'd done something Very Bad; Aleynikov knows a lot about high-speed trading systems; one thing an expert on that stuff might do that would be Very Bad would be to snoop on everyone else's transactions; so let's write a big long story that basically alleges that he did.)
So either our author has extra information that s/he isn't owning up to, or it's bullshit (and either wrong or just lucky).
The author's previous story alleges that Sarah Palin is addicted to amphetamines, on the grounds that she is quite thin and sometimes talks incoherently on camera.
Bullshit seems like the simplest explanation here.
Code is available here: https://github.com/phil8192/ob-analytics, just in case someone decides to grab a good chunk of tick-by-tick data (orderbook) and wants to go down the rabbit hole...
I can’t tell because the post source is hugged to death but I’m just speculating that this is utilizing sec form 4 data or similar sources? There can be some interesting signal there that you could incorporate into a trading strategy but that’s mostly semantic and another story, I totally agree with you broadly. It will be a minor miracle if there isn’t fraud that washes out with a market crisis of this scope, but I’ll be amazed if the grand majority of trading isn’t above board.
[1] Running cat on a csv file would usually just print a list of numbers, yes? Instead he gets ascii art with months on the vertical axis, making it seem like time going back and forth, and some random commentary in the right margin. I dunno but something about it screams "mockup". As does the minimalist, ultra-stylish AWS window.
Edit: also, if it were me who got incredible returns from trading at high speed on indian markets using only machine learning on historical data, I would definitely keep quiet about it. Just sayin'...
reply