[This is a guest post from a reader who currently works in an arbitrage development team. He wanted to clear up a few points about what "automated trading" is and isn't.]
Ah, taking a mid-day nap and waking up with extra money in your trading account… who wouldn’t want to make money while sleeping?
That promise of set-it-and-forget-it money draws lots of traders into the field and attracts computer science and engineering students who suddenly “discover their interest in finance.”
Only one problem: “automated trading” is far from automated cash flow, and you always need human intervention.
To find out why and to learn all about algorithmic trading, arbitrage and other forms of automated trading, read on.
Part of the problem stems from all the terms used to describe “computer-assisted trading”:
- “Trading machines”
- “High-frequency trading”
- “Black-box trading”
People use these interchangeably but are referring to different concepts – so let’s clear that up.
Algorithmic Trading vs. Trade Origination
Here’s the key question you need to ask:
- Is a human making the trading decisions and simply having a computer help with the execution, or is the machine handling the execution and making the trading decisions?
The first category – where the computer only assists with the execution – is called algorithmic trading.
The second category – where the computer actually makes decisions – can be called trade origination, although that is not a canonical name (there isn’t any that I’m aware of).
So now we’ve cleared up the first misconception about algorithmic trading.
The second misconception: that algorithmic trading is about making the most money possible.
It’s actually about losing as little money as possible and reducing your costs.
To illustrate this, let’s walk through an example of how an algorithmic trading system might work:
Let’s say that you’re a pension fund manager and you’ve decided to sell 1 million of Company X’s shares that you currently own.
Notice how you – the human – make the initial decision here based on your analysis.
You want to get the best price available on the exchange that Company X’s stock is traded on, and you know that the highest bid price – the highest price at which someone else is willing to buy the stock – is $20.
So ideally, when you sell those 1 million shares you will get $20 million.
But not quite.
The problem is volume – most likely, that bid order for $20 was for far less than 1 million shares; it might have been your next-door neighbor trading 20 shares in his E*Trade account.
If that’s the case, then you’ll end up selling only 20 shares for $20 – and the remaining 999,980 shares will go for whatever the next best price after $20 is.
If the highest anyone else is willing to pay is $10, then you might only get around $10 million from your 1 million shares rather than $20 million – even though the bid price was $20 according to your trading software.
This problem is called “limited liquidity” in trading circles – you receive less than the “paper value” of your position because there aren’t enough bid orders at the price you thought you were getting.
As a human, you could simply monitor the market all day long and be on the lookout for those $20 bid orders.
That’s extremely time-consuming and labor-intensive, but that was exactly what agency execution traders did in the “old days.”
And that’s why algorithmic trading was invented – to manage the trading process over an extended period of time and get as close to the “paper value” as possible.
A trading algorithm for this scenario might divide the order up into many smaller pieces – 1,000-share blocks rather than 1 million all at once – and execute them over the course of a day or longer.
This is one of the key reasons why algorithmic trading has become so popular: there’s a high upfront investment, but a single machine can replace tens of pure agency execution traders – so you start seeing huge cost savings once you’ve been up and running for awhile.
Algorithmic trading saves traders a lot of time and money, but there’s a small problem: you still have to make your own decisions.
To make the process truly automated (in theory), various systems to originate trades were created.
To give a concrete example of how these systems work, though, we’ll focus on just one for now: non-statistical arbitrage.
Arbitrage refers to buying and selling multiple securities at the same time in the hope of making a profit.
The simplest type is non-statistical, or deterministic, arbitrage, where you find and exploit price discrepancies between 2 or more securities whose prices should be related. Ideally (ignoring technical issues), this kind of arbitrage is risk-free.
(Statistical arbitrage, by contrast, deals with expected values of securities over the long-term. There’s no guarantee that the future will behave like the past and so this is not risk-free in any form.)
Here’s how you might apply non-statistical arbitrage, and then how a computer could make it more effective:
The S&P 500 index has a futures contract associated with it – that just promises to deliver the stocks at a certain point in the future and is traded on an exchange.
The underlying stocks of the S&P 500 trade on exchange as well, so you can take their prices, figure out how much it would cost to hold the stocks until the future expires, and based on that decide whether the future is a bargain or rip-off at the current price.
So let’s say you think the future is too expensive – it’s $1,000 but the underlying stocks are worth only $990 and it will cost you $5 to hold them until the future’s expiration.
You can then sell the future and buy the underlying stocks – you deliver the stocks when the future expires and then make a profit based on the difference between what you thought the future was worth and the higher price you sold it for.
Does That Actually Work?
This is a very simple example, and it would never work in real-life because everyone else is looking for the same price discrepancies.
And even if you’re Rain Man, it would still take at least a few seconds to spot this type of price discrepancy…
…which is where machines come in. They can spot arbitrage opportunities like this in milliseconds rather than seconds or minutes, and make trading decisions far more quickly than any human.
If you were creating an algorithm like this, you might program in the specific securities or trends to look for in the market and then give exact instructions on what to buy and what to sell when certain conditions are met.
These days algorithms have become far more advanced and go well beyond just looking at prices – some actually try to scan news stories to determine “sentiment” for or against a company and make trading decisions based on that.
So going back to the terms at the top of this article, what does each one actually mean?
- “Algos” – Short for “algorithms” – Could be either algorithmic trading or trade origination.
- “Trading machines” – Generally trade origination.
- “High-frequency trading” – A type of trade origination system where securities are held for milliseconds (or less) rather than hours or days.
- “Black-box trading” – Might refer to either algorithmic trading or trade origination.
Many of these terms could refer to either variety of “computer-assisted trading,” so you need to dig in and ask what’s really going on when you see them.
Time to Retire to the Beach?
So you have a trade origination system set up and you’re making a lot of money with no intervention or decision-making on your part… time to retire?
Not so fast.
All types of automated trading systems must be supervised, checked, and updated constantly.
Even if the software itself is correct and has no bugs, market conditions themselves can be a “bug.”
We saw this back in 2008 during the start of the financial crisis when hedge funds started blowing up – supposedly “once-in-a-lifetime” events started happening every day and breaking all the old algorithms.
So no matter how great your algorithm is, it will only be effective until the next crisis, the next “unusual market condition,” or until everyone else starts copying you.
The top banks have spent a small fortune developing trading algorithms, and the tens of millions of dollars (or more) you need for such technology puts it well out-of-reach of anyone small.
And then there’s the small matter that no software is ever bug-free – especially when the algorithm is new, you need a human to monitor it all the time.
So even if your new automated trading program is making bank, you might want to hold off on buying that beach bungalow.
For Further Learning
This was just intended as a brief overview of automated trading – if you want to find out more, get a copy of Algorithmic Trading and DMA: An Introduction to Direct Access Trading Strategies by Barry Johnson.
Good luck with your trading, and let us know if you ever do make it to the beach for early retirement.