WALL STREET VS AI: WHO OUTPERFORMS?

Artificial intelligence is shaking things up and revolutionizing various industries (and Wall Street is no exception).

The only thing that must be truly understood is whether machine learning-based models consistently outperform investment banks and hedge funds in the long run.

Well… It’s not like a competition. Financial institutions know that algorithms are very powerful tools, and they started to incorporate complicated models in the early days of technology to make decisions in irrational markets without falling into the “trap” of human bias and emotions.

Models, in this sense, are magical tools that investors and traders can leverage to systematize their decision-making framework and profit from market inefficiencies.

Remember the mantra: “Disagreements create markets.”

The most disciplined investors don’t rely too much on their experience, but what they really value are the “base rates” over long time horizons.

As Dan Kahneman suggested in Thinking, Fast and Slow, a big part of trading is luck. (That explains the meteoric rise of quantitative research in the investment industry.)

Artificial intelligence, thanks to the analysis of huge datasets and incredible computational power, gave birth to many different ways of managing assets and institutional portfolios (think about high-frequency trading).

The name seems scary at first… well, in a sense, it is!

HFT (its acronym) is based on a speed-of-light execution of trades (fraction of a second) to take advantage of very, very small price discrepancies (that can arise almost from nowhere). Yes, markets are inefficient…

Unfortunately, or maybe not, we are not machines. Hence, we can’t complete thousands of mathematical computations in milliseconds. Traders have no choice but to "outsource” this task to algorithms.

These models are simply unreachable for even the most brilliant minds on the Planet (even if they developed them at first). There’s just one problem…

Many firms in the financial industry knew well in advance that this was an unbelievable opportunity to make profits “at the second.” Thus, many competitors (call them banks or hedge funds) started to take advantage of the same technologies. (They didn’t do that on purpose…)

Every time a “convenient” trade opportunity was found by these computerized models, they were almost doing the same thing (executing the same trade).

The more models were trading leveraging high-frequency, the more the tiny “spreads” became smaller. Smaller spreads implied lower profits (but still with quite high risks, given the negatively skewed nature of this kind of “bet”).

To say it in simple words, the models were all replicating each other. The market “manipulation” while the “subprime bubble” was deflating caused a market drop of around 1000 points(!)

The “flash crash” highlighted the power of these algorithms: much faster and more efficient than humans.

Ah, almost forgot that: machine learning-based programs didn’t just trade relentlessly and instantaneously, but they also processed immense datasets to adapt and make better decisions.

Wanna hear a fun fact?

For high-frequency trading, fiber optic connections are, in many cases, too slow! Every trade seems to happen in a matter of nanoseconds! (Not a joke.)

This strategy was used even before all these things were discovered: it was called “arbitrage”. But in that case, traders were counting on their “gut feelings” and their macroeconomic experience to speculate on micro-discrepancies in the prices of assets.

But without resources with impressive computational capabilities, this way of taking advantage of market disagreements sometimes caused catastrophic collapses (due to a lack of accurate risk-management practices).

Betting on random hunches (even if they were formulated by genial minds) provoked the failure of Long Term Capital Management, around 20 years ago.

In such volatile situations, human beings panic and start thinking irrationally as we see our money saying to us, “Sayonara”.

What about computers? They don’t care… They remain calm and disciplined in their strategy. Models will never be biased, and they will never exhibit some “hidden noise” in their decisions that often clouds human judgment.

Fear, greed, the temptation to “overleverage” to regain immediately the money lost in a bad trade, and a too aggressive reinvestment of profits (the so-called “gambling with the house money” effect) are just a few examples of many nuances that are present in every investor’s psyche.

News and rumors aren’t backed by solid data. Just words.

Algorithms ignore them as they base their “research” purely on data and odds (aka base rates).

With recent breakthroughs in portfolio management, blockchain, and quant research, AI might just be the future of finance.

Stay tuned! 🙂

Presented by: The Neural Chaos