Cortical AI Just Tripled Index Returns

reading patterns in financial waves

“Extinction of old forms is the almost inevitable consequence of the production of new forms […]”

There exists a deleterious belief about the financial world:

That financial markets are somehow too wild and random for machines to understand — they are simply too emotional, too susceptible to the whims of innumerable variables to decode.

Thus humans are indispensable for clever trading, and the careers of traditional speculators are as stable as their starched collars.

Nothing is Truly Random

Earlier this month, new temporally-aware Cortical Learning AI (CL-AI) yielded 24% non-compounded annual returns trading a portfolio of the S&P e-Mini, Gold & Copper futures over four years.

This Cortical Learning AI digests real-time, highly random data streaming in from publicly-available price charts, predicts when prices will rise or fall, makes an average of one trade per day.

By most opinions, this sort of bizarrely strong performance shouldn’t be possible. There are volumes already written on why machine learning can never truly predict financial markets.

Yet here stands a mechanical heretic proclaiming victory over flesh-based technical analysis.

So how predictable are financial markets?

Predictable enough!

Humans are infinitely complex creatures, but there’s patterns to our behavior. Rules and heuristics govern our actions to a greater degree than we’d like to admit.

Keep in mind: We’re social animals. When in doubt (and life is full of doubt), we tend to follow the crowd.

If you shout “fire” in a crowded theater, you can bet people will rush to the exits. If the media features stories of toilet paper shortages, you can bet people will rush to hoard it.

We take behavioral clues from our context: We’re afraid of missing out. Evolution wrote FOMO into our deepest subconscious. We act in waves.

This makes us predictable — as long as you can see and understand the patterns and relationships.

Tidal Markets

Financial markets constantly roll in ever-changing flux, moving like the currents of a vast ocean. Prices swell and crest in great waves, buoying some investors to rich heights and crashing others to abyssal depths.

What determines these waves is terribly complex: investor opinion, government policy, the tides of economic conditions, currents of fear and greed and yes, tweets — as well as a multitude of other factors more numerous than all of the fish in the sea.

This vast complexity necessitates years of study and financial analysis for an investor to make a steady profit.
It’s normally quite safe to bet that no machine can succeed at this game, but Intelletic Trading Systems says their Cortical Learning AI can.

But how can they possibly predict the myriad outputs of price direction, magnitude and timing without a full understanding of the input variables that create them?

The same way oceanographers predict tsunamis before they hit.


You don’t need to know every variable about every cubic inch of saltwater at all times. Ocean buoys and coastal tide gauges measure and report changes in sea level, while seismograph stations keep watch for earthquakes.

NASA uses GPS software that measures minute changes in ground & ocean height, which accurately predicted the size of the wave following the 2010 Chilean earthquake.

These are temporally-aware systems that use past waves (sea levels) to predict future waves. Past outputs become inputs to predict future outputs.

Temporal Memory

I’ve previously written about Intelletic Trading Systems’ cortical AI “Price Prediction Alerts for Bitcoin, and they’ve recently published their futures portfolio model’s fairly sensational performance.

Intelletic’s 24% non-compounded annual returns are lurid enough for me to dive into their new trading platform & see how it compares with simply buying and holding.

Their Cortical AI is a proprietary hybrid model that builds upon Numenta’s Hierarchical Temporal Memory networks (HTM), a new neurologically-inspired architecture uniquely suited to learning sequences in time-series data.

I’ve written plenty about what makes HTM networks so interesting, but instead of going into binary neuron activation or local Hebbian learning, here’s the four most important points.


  1. Predict based on past distinct sequences/permutations (The output D from A -> B -> D is treated differently than D from A -> C -> D)
  2. Are incredibly resistant to noise (up to 40% or more)
  3. Learn continuously, predicting as they go (no batching or re-training needed)
  4. Can predict multiple simultaneous outputs (with different probabilities)

Intelletic’s Cortical AI makes great use of these qualities:

HTM architecture is modeled after the many-layered cortical columns within the neocortex

Financial markets have a low signal to noise ratio. They move too quickly — and too randomly — for conventional machine learning models, which need to stop and re-train with exponentially- expanding training sets.
A new name has even emerged for when machine learning training sets can never be large or representative enough to obtain consistently successful results — ‘data shift’.

Similar prices, without knowledge of the requisite inputs, can also yield multiple wildly different predictions — depending on the patterns that immediately preceded them.

Context-Based Predictions

No machine can say with certainty:
“X minutes from now, the price of a specific financial asset will be exactly $Y”

Intelletic’s Price Prediction Alerts craft an alternate solution.
PPAs are current, forward-looking, time-sensitive price change predictions with probabilities that learn from their previous predictions (both successes and failures).

The CL-AI constantly searches for temporal patterns and relationships in real-time, highly random streaming price data. Certain sequences have the potential to trigger an alert, dependent on other dynamic factors such as price magnitude and volatility ratios.

Even sparse information with only a small (but important) match between factors can raise an alert. Along the same lines, your brain is able to read subtle environmental clues to recognize situations you’ve seen before, or recognize familiar patterns in new experiences.

For example, the below Price Chart of the S&P e-Mini is plotted here with Intelletic’s Price Prediction Alerts underneath.
The green horizontal bars indicate “long” PPAs (predicting a price rise), and red horizontal bars indicate “short” PPAs (predicting a price fall).
The dotted vertical line denotes the “current time”, with the past to the left and the future to the right.

Notice that multiple Alert bars (long and short) of varying duration can be active within the same time periods: potential future outcomes overlap in time with varying probabilities.

This is quite normal for a market filled with random movement. Smaller downtrends may occur within larger uptrends, for example. Price Prediction Alerts are best understood not as a promise of full certainty, but informative, highly interpretable notices to assist a coherent, risk-aware trading strategy.

Confidence Intervals

The interpretability is a result of three components: direction, time period and confidence levels.

Intelletic’s Price Prediction Alerts currently identify 6 unique “scents” in the market: long (rising) or short (falling) prices, active over the next 1.25, 7.5 or 15 hours — represented by the length of the PPA bar respectively (see the above chart).

To deal with real-world complexity, a Price Prediction Alert shows four fixed probabilities for price movement: Confidence levels of 75%, 50%, 25% and 10% for a defined $ magnitude change based upon what actually occurred during historic Price Prediction Alert periods.

A “30-minute long” Price Prediction Alert with 75% confidence of $50 simply means:
“During 75% of previous 30-minute long alerts, the price rose by at least $50 at some point during its active time period”.

I liken it to a pig sniffing for truffles: there’s not a truffle under every tree, but when the pig starts digging you’d do well to pay attention.

Avoiding Black Boxes

We have to be certain that 1) Price Prediction Alerts are consistent and won’t mislead you in direction or time, and 2) the market actually tends to move as the PPAs predict.

The key here is asking “How large are the average price shifts predicted in PPAs?”.

If the model can’t clearly distinguish patterns in the market, then it won’t be able to predict large enough waves to consistently make a profit at >50% confidence.

Saying “During 75% of previous 1.25-hour short alerts, the market fell by at least $X illustrates that Price Prediction Alerts of the same direction & length are consistent with each other (>75% of the time) — so long as $X is large enough to make an acceptable profit relative to risk.

If $X is generally significant enough to look good on a graph (as we’ll see later, it does), then the market does indeed reflect alert predictions, and the model is externally consistent with reality — the market is “predictable enough”.

Even “good” AI usually faces the “Black Box” problem: The process of turning input data into output predictions generally isn’t human-understandable.

But even if it works, how interpretable or truly usable are its predictions?

Recall that each Price Prediction Alert deals out 75%, 50%, 25% and 10% confidence levels with increasingly greater price changes, as large waves happen less often than smaller waves.

This logic makes the model surprisingly interpretable relative to your personal trading strategy and risk tolerance.
You can bet comfortably on 75% odds — or gamble on a 50% or 25%, waiting for a larger threshold at which to enter or exit a trade.

You could also think of it as a “wave height prediction engine”: Based upon real heights of past waves & its own predictions, it sends out warnings when waves are likely to change in the near future.

Underlying all of this is the idea that specific, identifiable price patterns and market conditions precede respectively distinguishable waves.

This is supported by the fact that the model is constantly monitoring prices, but only selectively issues a Price Prediction Alert. Bitcoin’s price fluctuates every minute, 24/7 — but the Bitcoin version of the model only issued on average one PPA per day.

Smart Automated-Trading

The AI’s predictions are sound. But good predictions alone don’t assure victory.
You also need a thorough — and consistent — strategy to use the predictions in smart decisions.

To truly profit amidst market pricing noise and chaos, an investor requires two things:

  1. The ability to accurately predict the future price of a financial asset in a defined time with a defined probability of occurrence (Price Prediction Alerts)
  2. Dynamic results-driven trading rules for both entry and exit to act upon the predictions. This could be an experienced trader, an automated system or a hybrid of both

I know this well: I built a simple algorithm that traded virtual portfolios based on Intelletic’s Bitcoin Price Prediction Alerts from 2016 to 2020.


My robo-trader puts full trust in the PPAs, only acting when one arises. Let’s say it starts with $5000 USD and one BTC.

If a [15 hour long] Price Prediction Alert arose with a [75% confidence of $132 increase in BTC price], then the price rose by at least $132 during 75% of past alerts of this type.

Since the 75% is drawn from actual historic Price Prediction Alerts, the AI’s real “guess” is thus:

The market conditions behind the current Price Prediction Alert will lead to similar price movements as those of actual historic alerts of the same kind.

My algorithm then converts 10% of its $USD to BTC, and sets an auto-stop order to sell 10% of its BTC if the price rises by at least $132 during the alert’s active period.

This naïve Python script simply bets on how good the Cortical Learning AI’s nose is, but doesn’t pay attention to the market itself. It never considers how well its past trades had gone, nor does it calculate safety metrics or maximum drawdowns.

It worked better than I expected, to be quite honest. The 1,400 90-day portfolios I tested yielded an average of 9% returns, but I was unable to push annual returns much higher than 14%.

As I often remind myself: If you find yourself falling short, you’re probably missing nuance.

The Robot Day (and Night) Trader

If you’re looking to predict the height of future waves, you may wish to pay attention to the weather — and whether you’re sailing around the treacherous Cape Horn of Africa or the placid waters of the Adriatic. Context matters.

In this latest backtest, Intelletic used their Intelligent Trading Platform (ITP): A multi-layered, 100% automated trading platform that applies various strategies and risk management techniques to maximize the advantages that the Price Prediction Alerts offer.

robots are inevitably coming for (most of) our jobs

PPAs advise you on “when there may be a good trading opportunity”.

Intelletic’s Intelligent Trading Platform decides “if that opportunity is good enough, when and how to enter & exit trades” to best make or to protect profits and trading capital.

In the same way that a cruise ship’s autopilot automatically adjusts stabilizers, speed & yaw to adapt to changing marine and climate conditions, Intelletic’s ITP adapts the trading rules to maximize the odds of success while carefully managing risk. Rough seas require a steadier — and perhaps calmer — hand on the wheel.

The safety margins themselves are technically interesting. Instead of a simple “Bet X% of your portfolio on 75% chances”, the Platform considers multiple interlinked factors, such as how dramatically the market’s been moving lately.
In recent times of high volatility, such factors cannot be safely ignored.

Overall, the Platform assesses risk and adjusts its rules to safely accumulate many small gains, minimize losses when trades don’t go as predicted and capitalize on the occasional large gain.

Knowing ahead of time when and where huge waves and troughs may occur is a huge boon in navigating the rolling tides of financial markets.

Diversity is Strength

Intelletic’s Cortical Learning AI will work with any assets where real-time streaming data is available. They believe that there are at least 800 financial assets in five asset classes with potential.

Carefully combining multiple assets in a portfolio enables Intelletic to safely increase portfolio profitability through diversification.
By combining assets with negative correlations — assets that tend to move in somewhat opposite directions under the same market conditions — trading capital can be better utilized and overall returns will be smoother.

In the spirit of our ocean waves and ships conversation: Imagine you’re a 16th century financial backer of trading voyages to the New World of the Americas.

Would you feel more secure dividing your venture capital between five individual sailing ships and crews traveling to different locations in the New World? Or would you rather risk it all on one ship and crew going to one destination?

Let’s say on average that one ship usually sinks in stormy weather somewhere on the journey and another usually returns with a half full hold.

Without knowing each ship’s exact fate in advance, splitting your investment among five ships would still yield good returns. Depending entirely on one ship has a significant chance to leave your hopes at the bottom of the sea.

This is the power of a diversified portfolio of somewhat negatively correlated ships traveling to different destinations under variable seas.

Intelletic’s portfolio achieved a 24% non-compounded annual return holding only three somewhat negatively correlated futures assets — gold, copper and the S&P e-Mini — and achieved a higher portfolio return than any of the three individual assets would have returned on their own or as an equally weighted passive portfolio.

What’s interesting is that with only three futures assets, the Intelligent Trading Platform only employed the capital in trades about 6% of the time.

The other 94% of the time the trading capital was sitting in cash waiting for the right wind to blow and waves to form for gold, copper or S&P e-Mini trading opportunities.

With such sparse use of the trading capital there is little interference between trades of each of the three assets, leaving plenty of room to add other negatively-correlated assets to the portfolio (perhaps hog futures or Tesla).

Trading Results

Without further ado, here’s the backtesting results graph. From the bottom up, the first three coloured lines represent the accumulated profit for each of the three futures assets. Green is the aggregate profit of Gold, Copper and the S&P e-Mini.

The initial investment in the portfolio was $32,500, divided evenly into one contract for each of the three futures assets. Intelletic’s portfolio achieved $32,000 of profit in four years, about 24% non-compounded average annual returns.

But the shape of development is also important. Profit growth rises in the first year, then hits a plateau in the middle where profits maintain before continuing their rise later on.

Ideally, a portfolio’s profit curve should be more stable, rising at a constant 45 degrees from left to right.

What was occuring during the middle temporary plateau in accumulated profits?

Either 1) the Intelligent Trading Platform is not doing its job, or 2) there were not enough significant opportunities to profitably trade the three assets in the portfolio, so why risk trading?

Considering the fine profit made in 2016, the latter is more likely.

Time is money; thus idle money is missed potential profit.

Tepuis are a generally undesirable shape for your portfolio growth curve

With only three assets, Intelletic’s ITP is sitting in cash 94% of the time waiting for the right wave (or trough).

The clearest solution is to put the idle capital to work more often by adding additional trading assets to the portfolio:
If you’re not seeing enough opportunity, broaden your horizons.

Increasing asset diversity with more somewhat negatively correlated trading assets will generally improve the shape of profit aggregation, and should result in additional growth in the flatter middle part of the green aggregate profit curve.

When Gold, Copper and S&P e-Mini futures prices aren’t presenting suitable trading opportunities, other assets might be. The 94:6 ratio shows that there’s plenty of room to add more diverse assets without overlapping or interfering with the portfolio’s existing trades.

For example, Bitcoin spiked to nearly $20,000 in late 2017 & fluctuated wildly in early 2018, right when this portfolio wasn’t seeing much action with its three futures assets.
Incorporating more diverse assets such as cryptocurrencies would give Intelletic’s Intelligent Trading Platform much more room to capitalize on new waves of opportunity and further add to the annual return of the portfolio.


When faced with grand back-testing results, the immediate question to ask is whether or not it’s a case of “a rising tide lifts all boats”.

The ITP profited from both long and short opportunities, incrementally gaining from the crests and troughs of price waves.

Is Intelletic’s portfolio performing well because of prescient AI Alerts & a well-designed trading Platform?
Or is it prospering only because the chosen assets just happened to continually rise on their own?

How does it compare to simply buying & holding the same assets?

Compare these 3 graphs of each asset’s price during the portfolio’s timespan (click on “5Year” to see a comparable timeframe).

From March 2016 to January 2020:

Gold went from $1234 to $1528 | 24% increase (6% annual)
Copper went from $2.15 to $2.82 | 30% increase (10% annual)
E-mini S&P went from $1930 to $3259 | 68% increase (17% annual)

If we invested an equal sum in each of these three assets to build out our portfolio, and simply held for the comparable four year period, we’d see a weighted average annual return of 11% as the assets rose in the long term.

Intelletic’s Intelligent Trading Platform actively traded on both long and short waves, and won the race with an average annual non-compounded return of 24% during the same four turbulent years.

‘Disruptive’ is Overused. Call it ‘Progressive’

Make no mistake, Intelletic’s results are impressive — the 24% average annuals are non-compounded, so a compounded investment would have likely beat 30%.

But are their Cortical AI predictions and trading platform worth betting on?

Judging by their actions, Intelletic is happily placing their money where their collective mouth is. Their plans include a “Three Pillar Revenue Strategy”:

  1. Proprietary trading: using PPAs & the ITP to manage their own investments
  2. External private trading: remotely executing trades in privately-held accounts worldwide on a profit sharing basis
  3. SaaS: offering price predictions as a tool to online traders via online monthly subscription

They believe that their largest and most reliable source of income will be from raising further debt and equity capital to employ in their proprietary trading fund and allowing their Cortically-informed Intelligent Trading Platform to do what it does best — trade on an automated basis 24/5.

I find this fairly reassuring:

If someone loudly offers to sell you a magic elixir of health, but refuses to drink it themself, you might question the benefits of such a potion.

If they spend most of their time quietly guzzling the brew, leaving a small tray of samples out for passersby, it may be worth further investigation.

The Great Wheel of Growth Turns

The economist Joseph Schumpeter wrote of capitalism as a process of creative destruction:

The driving animus of capitalist economies is a “perennial gale” of constant movement and innovation, much as sharks must always keep swimming or suffocate.

Technology tends to change the world slowly, and then all at once.
The automobile was initially met with derision in New York City, until Henry Ford’s Model T drove horse carriages out of business. A new world in just a decade.

Online ad-based news sites herald the death of traditional newspaper,
Automated robots replace the jobs of factory laborers,
Surely the robots won’t come for our intellectual white-collar jobs next?

Technological innovation incessantly outperforms the old paradigms, and those who ignore such changes are inevitably trampled by those who rush to leverage the new opportunities.

The modern economy is death-set in its endless quest for efficiency and marginal profit, boats borne forth ceaselessly into the future.

In such a world, I can hardly doubt the emergence of Intelletic’s Cortical AI asset trading, or something much like it.

It would be quite arrogant to think that financial markets, finite systems of human creation, are immune to the intelligence of such machines as they grow further beyond our limited capabilities.

The question is not “will machines replace us” in this gale of creative destruction, only “when”.

data scientist, machine learning engineer. passionate about ecology, biotech and AI.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store