GRAYMAT Trading Systems

The ultimate cold blooded approach to market trading

System Trading, also known as algorithmic trading, commonly involves the use of computer programs for sending trading orders, with the computer algorithm taking decisions on time, price and volume of transactions. GRAYMAT systems are of the directional flavor : they generate Long/Short market orders.

Trading systems can be an excellent support for "classical" or "discretionary" money managers, but we strongly believe that they are best employed in automatic mode: it just lets the algorithm trade, without any human interaction.

It may seem counter-intuitive, but provided that the right money management scheme is in place, auto-mode notably reduces risks and, in the long run, increases returns.

These benefits are even more obvious in periods of crisis, when human traders accumulate large losses as a result of stress, whereas trading systems just produce normal losses and gains, or go out of the market. We are pleased to note that more and more investors show interest for this consistent and pragmatic approach of the market.

Working on diversity

In the past, financial mathematicians have brought rich contributions and theories enabling practitioners to extract useful information about the trend or reversals of the markets. They have legated a large set of indicators, patterns and algorithms that partially model the market.

More recently, applied mathematics in "Data Mining and Statistical Modeling" have brought new contributions for a more comprehensive view of the markets.

At GRAYMAT, we have developed a proprietary AI algorithm which creates candidate trading strategies from mixed market models covering the broadest range of possible statistical states.

This methodology brings two features:

• a single strategy is able to generate profits on several markets,
• different strategy structures can work favorably on the same market.

As a result, diversification is achieved both at the market level, and the system level.

Enhancing Confidence Levels

If trained inappropriately, a system can easily over-fit the data, which means that it has learnt the past too well (including noise), and that it will not perform satisfactorily in the future.

We have built in several features to reduce this risk:

• According to the Occam's razor principle, we keep our systems simple, a necessity when robustness is a concern.
• Then, we use a large number of systems in an ensemble which take collective decisions to further reduce risk and boost profits
• Finally, all our systems are validated on extensive unseen data and can only be released when they have passed the test of time.