BlackBox Machines™
is a NYC based FinTech startup that is developing algorithmic trading systems based on evolutionary AI models that can adapt to the increasingly chaotic and ever-changing financial markets.

From Gaming AI to Trading AI

AI is advancing rapidly. A new wave of AI algorithms are able to defeat humans in complex and dynamic games such as DOTA 2. Human players train for years before they master the game. Without providing any instructions about the game play and using only reward feedback, the latest algorithm by OpenAI learned the game by self-play. Only in 24 hours, it learned to beat the best professional human players consistently. Watch the video.

The AI's success was not based on speed but on its self-learned strategy. Players described the AI as "scary" and were astonished at how "human like" and "precise" its moves were.

The revolutionary ability of the latest AI algorithms to adapt to and exploit dynamically changing game environments and even strategically "outsmart" human players applies perfectly to trading the markets.

The Goal

BlackBox Machines™ was founded with the objective to build sophisticated next-generation trading technology using the latest AI algorithms that are able to learn the unique personality and nuances of each individual market and to self-evolve robust trading strategies. The goal is to build a delta neutral portfolio of diverse AI strategies that generate consistent profits with the ultimate goal to start an AI-based hedge fund.

Traditional Algorithmic Trading

● Rigid set of rules that do not change
● Created by humans by testing a few ideas
● Scripting language with if/then statements
● Fitted only to historical data
● Stops working when the market changes
BlackBox Machines™ Trading AI

● Constantly learning from its experience
● Genetically evolved by testing millions of ideas
● Neural net with neuroplasticity similar to brain
● Searches for new opportunities every day
● Self-correcting and able to adapt to changes

Technology Overview

BlackBox Machines’ trading method is based on "reading the tape" - a common style of successful pro traders. The AI analyzes order book flow, price and volume data streams, and it makes quick decisions on when to trade. It is based on the fact that news drives the order flow, and the order flow drives the price. During a catalyst event there are temporary imbalances between supply and demand that can be detected by experienced traders or AI.

Tape Analyst™ is a subsystem that “reads the tape.” It monitors the order book and ticker tape data stream and calculates multitude of variables that help the AI discover short term supply and demand inbalances. It calculates various ratios, distances, angles, momentum, and acceleration from order inflow and outflow, time, volume, price action, support and resistance levels, and creates a set of normalized values called “feature vector” that is fed as an input to the neural net. Data feed adapters are available for popular exchanges and markets.

Strategy Sculptor™ allows a user to guide and evolve a neural net model by using a genetic algorithm on a 96 CPU supercomputer. Starting with random neural connections and using the same evolutionary principles as in nature: natural selection, DNA crossover, mutation, speciation and survival of the fittest, a "brain structure" of the model is evolved. Millions of diverse trading models are backtested while natural selection favors the ones with low drawdown and high consistency of profits. Curvefit prevention methods help evolve generalized models that work on unseen data.

Cortex Trader™ is a subsystem that uses the evolved neural net to execute trades on an exchange using limit or market orders or combination of both. It calculates the optimal size of a trade using Kelly Criterion in order to maximize the logarithm of wealth while trying to minimize its impact on the market. It manages all positions, active orders, and stop losses.

Risk Mitigation

BlackBox Machines' models are based on frequent trading in relatively small size keeping the exposure low. The goal is to build delta neutral portfolio of AI strategies. Hence, a black swan event should not affect overall equity curve to a significant degree. Our systems are ready to exit immediately at the first sign of an unusual adverse move and they never hold positions when the market is closed. Additionally an independent trade envelope protection system and kill switches keep all systems within the expected operating range. Curve fitting of the model is monitored and prevented.

Partnership Opportunity

BlackBox Machines™ is a fintech startup company based in New York City, and we are currently looking for advisory board members, pro-traders, strategic partners, and angel investors.

Contact us at

Copyright © 2021 BlackBox Machines. All rights reserved.