Machine Learning Meets Market Structure.

Space Machine’s goal is to deliver consistently strong risk-adjusted returns, regardless of market conditions and uncorrelated to any other asset classes, while keeping drawdown risks below acceptable thresholds.

The company extends leading academic research by conducting empirical studies of market structure using fine-grained financial time-series data, and applies state-of-the-art machine learning techniques to develop custom models and trading strategies.

Space Machine’s proprietary data analytics algorithms readily capture the underlying “rhythms and rhymes” of certain identifiable market regimes based on the timing of economic news releases, which are then used as trading signals for automated strategies to statistically extract expected profits by executing a large number of trades in the foreign exchange market.

In times of market uncertainty, Space Machine reverts to its Zeroth Rule of Trading:

Our Story

Space Machine was founded in 1999 by John F.C. Cheong, a U.C. Berkeley computer scientist. During its first decade, Space Machine was in the business of developing embedded navigation software for use in car navigation systems in collaboration with tier-1 automotive OEM partner.

Navigation products embedded with Space Machine’s software had won the 2007 CES Best of Innovations Award and the 2010 CES Innovations Design and Engineering Award. Space Machine’s navigation software has been selected for adoption in North America by Toyota and Tesla. Space Machine subsequently extended the geographic reach of its software into PND products for the Japan car navigation market starting in 2009.

By 2013, however, the competitive landscape made it quite clear that the company’s core navigation business based on the venerable software-licensing model had finally run its course. The time had come to evolve the business in a new direction.

Finding no easy pivots, a two-person tag team at Space Machine took the proverbial “leap into the deep end” and embarked on a quest for survival.

From their respective vantage points in Silicon Valley and Tokyo, the duo began to investigate an idea that has certain intuitive appeal to anyone who had worked extensively with map databases and routing algorithms, i.e., a new approach to quantitative trading based on distance latencies inherent in geographically dispersed trading venues.

With distance latencies upwards of 120+ milliseconds, the decentralized nature of the global foreign exchange market allows certain market inefficiencies to be readily observed and learned. In time, a bare-bones bootstrapped foreign exchange trading desk, built upon a makeshift assembly of open-source statistical software packages plus an off-the-shelf trading platform augmented with real-time data feed; began gathering form to take shape at the outer fringes of the financial cloud.

Why Forex?

Space Machine’s core technical expertise is in machine learning heuristics, real-time sensor data fusion, and high-performance databases. The foreign exchange market, because of its open hours and liquidity, is ideally suited for systematic parameter optimization from a machine learning perspective.

The Forex market normally operates 24 hours a day and only closes on weekends. What this means is that there are almost never any huge price gaps between trading days, as is common in stock markets. Continuity is good for machine learning algorithms.

Moreover, high liquidity means constant movements. For traders this means constant opportunities to make profitable trades, but for machines this means high-frequency market data for accurate regression tests. For professional traders, high liquidity also means that they can sell and buy larger amounts of assets at actual prices without the risk of moving the market prices.

Even better yet, the very small transaction costs and high leverage in foreign exchange make it possible to generate high profits on very small price movements. While such high leverage comes with its own risks, machines are far better than humans to precisely compute such risks and place properly-sized bets in quick response to information extracted from real-time market data feeds, or to get out of a position quickly when the market turns.

Despite its characterization of being a zero-sum trading game with negative expectations, foreign exchange remains an attractive first market for applying machine learning techniques to building systems and strategies for automated trading. Thus it is from this new way station that Space Machine’s journey began again.