The research and development effort that ultimately led to the creation of the I-System started in the summer of 1997. Alex Krainer, the founder of Krainer Analytics assembled a small team whose assignment was to devise a systematic approach to managing commodity price risk for a Swiss-based commodities trading company.
The team essentially began from a blank sheet of paper, researching broadly into any school of thought dealing with time-series analysis: study of cycles, Elliott Wave theory, autocorrelations, ARIMA, Fourier Analysis, Box-Jenkins method, fractals analysis and other mathematical approaches including technical analysis.
As the circumstances had it, our thinking converged upon trend following even as we were entirely unaware that systematic trend following already had a rich history and acceptance in the alternative asset management industry.
How we stumbled upon trend following
As we began our research during the late 1990s, global economic growth was in full swing and the demand for oil was rising. Meanwhile, funding for oil production and refining tightened globally as capital favored investments in information and telecommunications technologies.
We fully expected that demand would progressively outstrip supply and push oil prices significantly higher. Contrary to our expectations, oil prices more than halved from around $24/barrel in 1997 to below $10/barrel in 1999. In early 2000 the dotcom stock bubble burst and a recession set in.
Bearish market expectations were prevalent, and many traders thought that oil could halve again to $5/barrel. But again the markets defied our thinking: after bottoming out in 1999 oil prices tripled to $35/barrel over the following 20 months.
Trends lead, market narratives follow
It became apparent to us that prices were simply manifesting trends and that analyst explanations and market narratives followed the lead set by prices. This realization gave direction to our research and we set out to explicitly formulate the way we should determine market trends.
How to recognize price trends: exploring fuzzy logic and neural networks
The next problem to resolve was significantly more complex, but also infinitely more interesting: how do we recognize price trends? For a human analyst, the task is relatively easy – our brain is equipped with sophisticated organs of visual analysis and algorithms that do the job effortlessly. But to build a computer model that could do the same thing seemed difficult in the extreme. Part of the problem is that we can’t describe a trend with a crisp definition that could distinguish all trends from all non-trends. This is because a trend is a fuzzy concept where some patterns look more like trends than others. Modelling fuzzy concepts requires fuzzy logic and neural networks. This suggested the make-up of the core architecture of our model.
Determining trade entry and exit signals
Once we were able to determine price trends, we thought we could improve trading results by adding a set of algorithms that would define additional entry and exit signals for trades, including stop-loss trades when there were significant corrections in a trend as well as profit-taking trades where trends had made a significant advance. Here again we borrowed a number of studies from technical analysis including Stochastics, Relative Strength Index (RSI), Parabolic Stop-and-Reverse (SAR), Bollinger Bands, trendlines as well as algorithms that performed reversal pattern recognition (double tops and bottoms and head-and-shoulder patterns). Thus, in an up-trend, entry signals would produce buy decisions and exit signals would produce sell decisions. Exit signals could be either stop-loss or profit-taking trades. In a down-trend, entry signals generated sell decisions and exit signals the decisions to buy.
Building the proto-type (1999)
By the summer of 1999 we completed a prototype of the I-System. While we were very pleased with the solution we had built and its functionality, one disconcerting thing became apparent. The model’s complexity made it quite a challenge to maintain. Any error we found and corrected and every improvement we tried to effect tended to introduce multiple new errors.
We were eager to present our model to the management and to ask for funds to test it and begin live trading. At the same time we understood that our ability to develop the model’s full potential made it imperative that we improve its quality and reliability. Rather than asking funds to trade we asked funds to hire top notch software engineering talent and rebuild the model from the ground up.
We did this twice over, in fact. We ran the original prototype and two new versions in parallel until we were sure the model was bug-free and that the model did what we intended for it to do. This work took full four years.
Completing the industrial strength I-System (2003)
The first ‘industrial strength’ version of the I-System was completed in 2003. Over the next three years we made changes to the database architecture, improved the system’s ‘ease of use’ ergonomics and optimized the calculation processes, increasing the model’s speed by more than 40 times. When formulating trading strategies requires hundreds of thousands of iterations, the difference between 1.5 seconds and 0.04 seconds per backtest is absolutely massive. These changes vastly improved the system’s fidelity and ability to generate more high quality intelligent trading strategies.
More about the I-System:
Mastering uncertainty…
A very detailed account of the conceptual, philosophical and practical evolution that led to the creation of the I-System as well as our experiences using it is available in Alex Krainer’s book “Mastering Uncertainty in Commodities Trading” (2016), which was rated #1 book on commodities for investors and traders for 2021 and 2022 by FinancialExpert.co.uk. A PDF copy is available at the following link: https://isystemtf.files.wordpress.com/2021/11/masteringriskanduncertaintyincommoditiestrading_6x9printlayout.pdf
I-System Trend Following 