Hacker News Reader

Comment by mch82 | Hacker News Reader

My only issue with this article is that it seems to be written backwards. It's not a problem that Tesla published a joke on April Fools' Day, which is pretty much a trend in the tech industry since Google started doing it years back. The problem is that automated trading algorithms overreact to human behavior and cause human consequences. If anything, additional SEC scrutiny should target the algorithms and their corporate users. When jokes are illegal and harmful algorithms are legal, then culture and humanity are sucked out of life.

View on HN

> When jokes are illegal and harmful algorithms are legal

It worries me when even people on HN are talking about banning 'harmful' algorithms.

The SEC shouldn't get involved. The people whose algorithms ran up the stock and lost a bunch of money over a joke will learn their lesson, or will go out of business, soon enough.

It's the SECs job to regulate securities and exchanges. They regulate people, they can regulate algorithms.

It's not like Congress is passing a law that regulates what encryption standards can and cannot be used. It's the SEC regulating trade on the stock market no matter who or what is conducting the trades. Whether they should or should not ban it, it's well within their authority.

You're trying to make it into an all-computing type of issue. It's not. There is no slope, and it's certainly not slippery.

The SEC can get involved. The question is whether they should. See, people who write "bad" algorithms already get punished by the market and lose money. They already have an interest in improving.

What would a regulation against "bad" algorithms even look like?

They should be forced to give some of those losses back.

? For the most part, they tend to be the ones who lose.. so they've already given them back.

They did, before the end of the session.

"Giving back losses" was supposed to be humorous. Apologies.

Simple. All trades have a random time from 0-10 seconds prepended as a delay before the exchange starts the process to execute the trade.

Suddenly, microsecond trading algorithms are dead.

Nope. That existed for many years in implementation details at exchanges and the games were worse.

What actually happened in those cases is liquidity went away faster in drastic moves and the spreads had to account for the risk.

Further, you'd have all manner of snake oil consultants that would be selling true secrets of the NASDAQ randomizer (this exact thing happened before the exchanges standardized colocation. Guys would claim only they knew where in a data center had the shortest cable runs for instance). All of that inefficiency gets priced into the spread and is paid by all market participants.

If you really want to get rid of microsecond trading (I don't know why you'd care honestly, but if you did) get rid of the sub-penny rule. Make market making algorithms truly compete on price. Speed will still be important for canceling orders/gathering information, but you could counteract it by trying to do what we actually want the algos to do. Discover the right price cheaply.

No, this is wrong. You've simply increased the complexity and hence the risk of submitting a bid/ask which is going to lead to increased spreads and increased costs for people who buy and sell stock.

Um, so?

The goal here is to cut down on automated meltdowns caused by people trying to get a microsecond jump on everybody else.

A stock price is supposed to have some relation to the value of the company. With HFT, it is pretty clear that we left that realm many moons ago.

A random time within 10 seconds is hardly going to affect the stock price significantly. The fact that stock is going to be a couple of pennies more expensive isn't exactly going to gather any sympathy from me. And, a bigger bid/ask spread will stop stock price meltdowns which are more momentum-based (aka trader driven) rather than actual real information (aka business driven).

> See, people who write "bad" algorithms already get punished by the market and lose money.

You've misunderstood what is meant by "bad" algorithms. They aren't universally "bad". The vast majority of the time they make a ton of money. The problem is, the reason they make money is because they have limited safe guards that protect it from doing something stupid 99% of the time, but the remaining 1% is left a mess in order to ensure that 99% of the time they execute faster than the next guy. Unfortunately, they are an increasingly dominant force during that 1% window because they are a few nanoseconds faster to make a trade the rest of the time.

> What would a regulation against "bad" algorithms even look like?

You don't regulate the algorithm, just like you don't regulate against bad traders. What you do is set the rules so that the risks & rewards favour a healthy ecosystem. If sub millisecond trading yielded no benefit, it'd do a lot to prevent problems.

> It worries me when even people on HN are talking about banning 'harmful' algorithms.

Why? Algorithms can be harmful as much as anything else.

> The people whose algorithms ran up the stock and lost a bunch of money over a joke will learn their lesson, or will go out of business, soon enough.

The problem is that won't happen. The advantage of trading stupider, but slightly faster, than the other guy is too huge. Sure, sometimes you lose money, but most of the time you make money. It's classic game theory: the stupider but faster algorithm benefits from the slower but smarter algorithms... unfortunately though, it damages the ecosystem, and if it is allowed to continue it might actually grow to kill the ecosystem. While in theory, it is improving the liquidity of the system, the way the game is rigged, the rewards for the improved liquidity far outweigh the benefits.

> The SEC shouldn't get involved.

I'd argue quite the contrary. Now, the SEC shouldn't be saying, "you can use this algorithm" and "you can't use that algorithm", but I think there is a case to be made for imposing trading rules that don't reward a 5% speed improvement over 100ns equally to a 5% speed improvement over 100s.

None of what you said makes any sense. I can trivially implement a trading algorithm that is faster than anything else by ignoring the outside world and choosing a price at random.

Algorithmic traders don't benefit by simply being faster, they benefit by being faster to the correct price.

"Correct price" is not boolean. Speed and accuracy both have degrees and they trade off. High frequency traders use algorithms which are not as accurate as possible, but are fast. They knowingly use algorithms which have compromised accuracy for speed.

They therefore deserve zero sympathy when their algorithms fuck up. They knew what they were getting into, so they should accept the consequences with grace.

But the context wasn't people saying we shouldn't have sympathy, but that they should be illegal.

My two cents are that for something to be illegal there has to be a victim, and the victim can't be yourself. Whomever employed the persons responsible might have a legitimate claim to damages. But as they say in the kitchens I grew up working in, the traders eat their mistake.

I didn't realize anyone was suggesting they should be illegal. I was suggesting that the structure of the marketplaces have resulted in a flawed set of incentives that threaten the health of the market. There is a need to change the rules to keep the market healthy.

I certainly agree that when their algorithms fuck up they deserve no sympathy. They lost money, and that's that.

But that shouldn't be a reason to consider making operating on those algorithms illegal.

Can you see why this is a complete strawman? You just claimed that, because a completely random algorithm would be "fast" but incorrect, then there is no tradeoff between being fast and being correct.

What cbsmith makes complete sense:

    The advantage of trading stupider, but slightly faster, than the other guy is too huge.

This does not mean choosing at random, it means we are greatly advantaging simplistic algorithms that are just barely good enough.

The fitness landscape for these algorithms hasn't been designed by computer scientists, it just kinda happened through circumstance. As society gets more computerized, this sort of explicit design of computational spaces will become increasingly important.

I think that you and cbsmith significantly underestimate the sophistication of these algorithms. If you are so certain that they are dumb and arriving at incorrect prices (but quickly) it should be easy for you to go trade against them and make a lot of money.

"dumb" and "sophisticated" aren't mutually exclusive, especially when you look at this particular example: only sophisticated algorithms could automatically react to Tesla announcing new product, so only them could do dumb decision and take April Fools joke into account when trading.

The point they seem to be making is that benefits of fast trading massively outweigh the losses from making dumb decisions like there, so algorithms that integrate more and more input signals into decision making win in long term - no matter how rough are their assumptions about those signals.

It's not about algorithms being fast and simple - it's about algorithms doing more and more complicated stuff (not in "how", but "what" they do), so they can easily make (and will make more often with time) dumb decisions when looking from human point of view - but at the same time they can quickly react on much more types of events that normally would have to be noticed by human first, or already deduced from market actions.

> If you are so certain that they are dumb and arriving at incorrect prices (but quickly) it should be easy for you to go trade against them and make a lot of money.

It is easy to do this, as demonstrated by the article. Just con the algorithms with fake newsfeeds.

That's also illegal though, so no one is going to go out and prove it. They shouldn't need to when we can see natural experiments like this.

It's easy to con the algorithms now with a fake news story because such fake news stories are extremely rare and not worth programming against. If such stories were common the people who run these programs would update the to better reflect how the real world was working and your ability to profit against them using this strategy would go away.

> "I think that you and cbsmith significantly underestimate the sophistication of these algorithms."

Let's remember that this discussion was kicked off by some of these algorithms being tricked by a blatant April Fools joke.

The way some people talk about trading algorithms, you would think that they believe they are Strong AI that is lurking in datacenters biding their time. Really though, compared to a human, they are quite dumb. They just happen to be fast. Being smarter than the algorithm isn't sufficient if it is somewhat smart but blazing fast.

> Being smarter than the algorithm isn't sufficient if it is somewhat smart but blazing fast.

Sure it is. There are lots of people that make money in the stock market by making long term bets.

By that logic, this does happen: "it should be easy for you to go trade against them and make a lot of money."

So what the hell was your point?

Stock markets aren't zero-sum. You can successfully do long-term investments without having to "trade against them".

Which suggests that:

> "There are lots of people that make money in the stock market by making long term bets."

Is not actually a counter to:

> "Being smarter than the algorithm isn't sufficient [to trade against it] if it is somewhat smart but blazing fast."

Yes, there are a lot of buy side strategies, but what we're talking about here are HFT algorithms that balance out several times a second. They drive much of the liquidity of the markets and are the ones that react instantly to new pieces of information.

I think you grossly overestimate their sophistication. Shaving a few instructions off is helpful, avoiding cache misses is critical. There literally isn't enough computation time to do anything terribly sophisticated in terms of looking through large amounts of data to arrive at a conclusion about possible linguistic subtext.

Anyone else who happened to buy at that time was also affected.

There are always less-informed bystanders, and an efficient market is supposed to be about getting prices right. Adding noise to the signal isn't helpful.

I agree, though, that it will probably self-correct.

Then we should also punish those less-informed bystanders, since they're compounding the noise.

Well, they are "noise traders" but if they're not coordinated by a particular signal, on average it should balance out.

In any case, I don't really care much about punishment. I do care about fixing things.

Except that harmful algorithms aren't limited to only harming other algorithms. They manipulate the stock price for every investor, including small investors. The SEC should be involved to protect the small investors from the effects of bad actors.

"Manipulate" is a just a loaded term for influence, which is what every investor does by buying and selling.

If one buys shares because some other person's algorithm pushed them up, isn't (s)he gambling just like them? Why should (s)he be protected?

> "Manipulate" is a just a loaded term for influence, which is what every investor does by buying and selling.

True, but remember that the degree of influence can vary by several orders of magnitude.

> The SEC shouldn't get involved. The people whose algorithms ran up the stock and lost a bunch of money

What if the firm which loses money has cascading effect on the market?

Agreed. This submission is titled "Tesla Stockholders Can't Take a Joke". Although technically true, the misleading part is the stockholders being talked about are algorithms, not people.

We should be worrying not that Tesla released a joke PR that moved the price, but that we live in such an apparently unstable financial world where in the quest for milliseconds, HFT algorithms create non-negligable price swings in response to completely de-contextualized information.

Five years ago this wouldn't have been a blip on the radar. If the rules have changed, so be it, but we should recognize that it's the system we've built that has changed the rules. We can yell 'securities fraud' at Tesla's PR department, but if so, we should recognize it's fraud because Tesla was releasing information that confused the dominant traders, which are now algorithms. If that's fraud, do we now have a fiduciary duty towards the algorithms?

More troubling, if the system we've built is so unstable around just PR releases, what's going to happen next time there's a serious problematic signal?

> Although technically true, the misleading part is the stockholders being talked about are algorithms, not people.

That was on purpose though. The title was meant to be deliberately misleading to highlight the fact that what we think of as "stockholders" is wrong: increasingly, the "stockholder" is an algorithm.

The price swing was negligible.

I don't think the SEC should get involved at all. The Tesla thing was obviously a joke to any reasonable human, so that's not a regulatory issue.

And if some reasonable humans choose to employ algorithmic trading systems, and then as a result lose some money because the algorithm wasn't properly tuned, that's not a regulatory issue either. People lose money on stocks because of bad decisions all the time.

Only possible role I can think of would be if there was some additional evidence (like an email chain or something) that Tesla did this release specifically to cause this stock movement. Seems unlikely.

Yeah, I can't say that I have any sympathy for those who have chosen to attempt to automate making money in this fashion.

> Only possible role I can think of would be if there was some additional evidence (like an email chain or something) that Tesla did this release specifically to cause this stock movement. Seems unlikely.

What if Tesla has press releases that aren't human readable but trolls automated trading systems? Like a spam trap but for trading?

Yeah, the phrase "shareholder-algorithms" stopped me cold.

It seems that people lost because they were either speculating blindly or had an algorithm that was working poorly. I don't see reason for these algorithms to be outlawed.

> My only issue with this article is that it seems to be written backwards.

I think you should reread the article. It is pointing out that the problem is the automated trading algorithms (and most importantly, how those algorithms prioritize expediency over intelligence). It points out that what Tesla did wouldn't be a problem except for the fact that we've got computerized traders.

It would be really interesting to check if other companies who put out April Fool's press releases impacted their stock prices as well.

Reply on news.ycombinator.com