Supercomputer predicts next month’s stock prices?

If you’re anything like most traders, then you know what it’s like to lose money.

The journey from pathetic to profitable for me was chalk full of nonsense, idiocy, false gods, and outright fraud.

Like many, I started watching the financial news with their money honeys and opinionated imbeciles.

I would often say to myself: “Self, how in the world can there be 10 wildly different opinions on the same stock”?

Who would allow these guys to manage millions of dollars? I wouldn’t trust them to make my sandwich.

Kaboom! That’s the sound of my first trading account going to zero – with money I had borrowed at 20% interest.

If I could travel back in time in my DeLorean, I would cut that card up and scatter the pieces to the wind.

I later found out that 80% of mutual fund managers regularly under-perform the S&P 500 index.

So I bought a book on technical analysis.

The idea that price patterns have predictive powers sounded right up my alley as a more analytical thinker.

But after studying Elliott Wave, wedges, head and shoulders…I still lost money.

Give the same chart to 10 chartists, and you’ll get 10 different predictions.

That’s pseudo-science in action.

Kaboom! That’s the sound of my second trading account imploding. Back to square one…

Around that time I was learning how to count cards at blackjack.

Notice I said “learning.” I certainly wasn't making money at it.

It was a simple high-low system, and I knew my odds only went up 1% (or $1 for every $100 bet I make).

And that’s if I didn’t suck down their “free” cocktails.

If you simply play basic strategy at a regular casino, you can expect to lose around 50 cents for every $100 bet you make.

Most people don’t even know basic strategy, so they lose $1-$2-$3-$4 or more per hand on average.

I've seen a guy so drunk he could barely sit in his seat hit on 20.

Sounds like my trading when I first started.

What playing blackjack made me do was realize that I needed to know my exact (or as close to exact as I could get) odds for every trade I made.

Otherwise, you’ll lose money faster than bad news traveling at a church social.

For example, I know that the average ‘DB Transaction’ trade makes $3.53 for every $100 since 1988.

Which gets to my next evolution as a trader: How do you figure out those odds?

The answer of course is back testing. You put some rules into a computer, and see what would have happened over time.

I’ve been a billionaire many times over in my back testing, so obviously there are some things that can go REALLY wrong when it comes to real world trading.

The most notorious problem is over-fitting the trading rules to the data.

When it comes to real world trading, the wheels fall off your bus on the way to the cover of Fortune Magazine.

Many of us “tested analysis” masochists set out to be trading swashbucklers, but we end up buckling before we swash.

The problem is systemic.

Most of the back testing software out there has a serious flaw – they allow you to optimize your trading rules on one stock, ETF, or commodity at a time.

Most would-be scientific-minded traders didn’t take (or don’t remember) statistics, and the perils of low statistical significance rise into existence like a demon conjured from the pits of hell.

So you think you have a diamond of a trading strategy, when all you have is a hunk of broken glass.

You need to have thousands of trades, which means you have to test your rules over hundreds of stocks (or more).

I’ve been misled by more than one company – names you would know – by presenting strategies with over-fit results.

They would tout how their system made 550% over the past two years, but if you use those same genocidal rules over a 20-year period, the results would scare a drunk man sober.

So the fix is to use the same formula on hundreds of stocks over decades of data.

Make sure to leave out 20% of the data while you’re optimizing. If that unseen data results in poor performance, then the odds of your system making you money in the real world are near zero.

Might as well piss your money away in Vegas and get some free drinks.

While I’ve simplified things a bit for brevity’s sake, what I just described has been the framework for most of my winning trading strategies.

But what I want to talk about in this blog post has to do with my latest evolution as a trader.

Knowing that a strategy is going to win 60% of the time, and make me 3.5% on average is great – but I want to go a step further…

I want to forecast the (nearly) exact price every stock will trade at a month from now, and only trade those stocks predicted to go up the most.

That would be like playing blackjack with all the cards facing up right?

Imagine being able to say “well, I looked at the entire NASDAQ 100, and I predict the top gainers will be Apple up 7.5%, Tesla up 6.3%, Vertex up 6.1%, and Amazon up 5.9%. I’ll trade those four, forget the rest, and go play a round of golf.”

Sound like science fiction?

I’ve been working on and off this project for the past 17 years. Like my early years of trading, I started with failure, and now am looking at something much more tangible.

It wasn’t until I learned how the music app called Shazam works to identify any song being played from a 10 second clip that I finally had a major break-through.

Then we went out and purchased the most complete stock database with nearly 400,000 years worth of price history on tens of thousands of stocks (including those that have been delisted over the decades).

Over the coming months, I’m going to write about what we discover. For better or worse.

I call it ‘Project 88’ as an homage to Back to the Future where Doc Brown and Marty McFly must take Doc's DeLorean to 88 MPH to travel through time.

You might remember in Back to the Future how Biff gets his greedy hands on an almanac from the future, and proceeds to make a fortune from betting on the outcome of a variety of sporting events.

That’s sounds kinda like what I’m trying to do with Project 88 right?

As you can imagine with nearly 400,000 years of historical data, a gargantuan amount of processing power is required.

Currently, we’re renting a bunch of cores on Microsoft’s cloud service to tackle that problem.

My lead programmer also made a breakthrough in pattern recognition speed, which is the secret behind the price forecasts.

After playing with some numbers, my first takeaway is that if you asked me how to make Elliott Wave books better, I’d tell you to burn them. At least then you'd get something outta them on a cold night.

And those fill in the blank XYZ patterns? Hogwash.

We’ve analyzed every conceivable price pattern in nearly 400,000 years of history and 50,000 instruments, and the patterns hardly repeat if you are too restrictive with your description.

So you have to be a little more vague with your description of the pattern.

The early pioneers of OCR (optical character recognition) had to do the same thing. The letter “A” can have many different fonts…or the paper you’re scanning into the computer can be tilted. So you have to account for that.

I realize I’m way over 1000 words into this post, so let’s continue the discussion next week.

Until then, here’s a quick 1 min 12 second overview of how Shazam works:

Continue reading Part 2

Trade smart,

Dan “Prince of Proof” Murphy