FREQUENTLY ASKED QUESTIONS

What is CONCERT AI?

Since 1988, Ferrell Capital Management has applied methodologies of risk management to successfully manage investments.  This philosophy incorporates risk-adjusted portfolio allocation as well as utilizing cash as an asset class to provide a safe harbor in difficult market conditions.  This logically evolved into the Ferrell CONCERT funds, and has provided daily optimizations to portfolios to achieve the above goals.

CONCERT AI takes this methodology to the next level though the use of Artificial Intelligence and Neural Networks that are capable of learning and self-optimization.  This provides the advantages of risk-adjusted portfolio allocation, cash as a safe harbor to prevent loss during market downturns, and constant self-optimization of the process.

 

What makes CONCERT a better way to invest?

With CONCERT, your investment will always favor the optimal risk adjusted markets.  As an historical example, consider Emerging Market indices and the large Exchange Traded Funds (ETFs) that track them, such as EFA & SPY.  CONCERT provided diversification benefits -- however, exposures changed as market risks and prices changed over time. The best strategy at some times was to exit the ETF for a country entirely.  CONCERT's goal was always to maximize the upside, eliminating components that could erase profits by parking the assets in cash until it was statistically safe to resume utilization of the component.  This philosophy is utilized for CONCERT AI as well, but with the significant advantage of continuous improvement and optimization through machine learning algorithms.

 

Managing investment through risk is transformational.  How big is the idea and what will be the impact on the industry?

Investors care about investment risk first and foremost.  To determine a client's risk tolerance - we ask how much the client is willing to lose. With CONCERT, the industry finally has a methodology addressing this most important priority: risk...

Markets and the risk profile of investments change daily.  Static portfolios do not adjust - they simply take a beating when environments becomes hostile.  Investing doesn't have to be done this way. Technology enables us to understand and track market characteristics at a much deeper level. We are tactical with investment exposures to stay in tune with the times. To use an analogy, think of the difference between using a Farmers' Almanac to predict the weather versus the dynamic display of Doppler Radar. We don't rely upon short, medium or long-term forecasts to determine our allocations.  Instead, our decisions are dependent upon daily risk-centric monitoring, facilitating an agile and adaptive methodology.

 

What does "Stay in Tune" mean?

This means adapting to the ebb and flow of market dynamics.   We find the best risk-adjusted returns at the moment, which greatly enhances overall portfolio performance, and provides what investors want - capital preservation.

 

What happens when investments turn sour?

The client's portfolio is quickly re-balanced to minimize negative returns. Assets are allocated to better opportunities, or if none exist, to cash.  Our algorithms "ride out the storm" - then get back in the game at the optimal time to resume aggressive growth.

 

What is different about how you manage money?

We do not make forecasts about fundamental factors, such as earnings, or the direction of investment markets.  Instead, on a daily basis, we interpret risk and make trading decisions based upon volatility, correlation, pace of change, and other selected statistics.  Ferrell Capital Management LLC performs fundamental risk analysis.  We believe our methodology is revolutionary because it challenges widely held beliefs and incorporates data overlooked by most managers.  We think differently...

 

Do you diversify by using bonds and cash?

These are the traditional means of reducing or managing risk.  However, it can be shown that diversification doesn't work when markets become highly correlated or the relationships between markets change.  And today, with low interest rates, an investor needs an alternative to safe-havens with zero or negative real returns.  We will use bonds and cash but only when our model cannot find more attractive risk-adjusted equity investments.

 

Is CONCERT a strategy or methodology?

CONCERT is a methodology.  We can apply our methodology to any asset class - as long as ample dispersion exists between the investment components.  Thus CONCERT can be applied to many asset classes, but not to all. (As an example, a U.S. Treasury strategy does not make sense due to the low volatility and high correlation of the market). 

 

Why do you use the SPY rather than the SPTR as a benchmark for comparison with CONCERT AI performance?

All our return calculations are based purely on price action and pattern recognition and do not include any dividend distribution. Our process is driven by absolute equity price performance as opposed to total return, knowing that over time increase (decreases) in dividend payouts will ultimately be reflected in the equity price. Accordingly, were we to use SPTR, we would not necessarily be comparing apples to apples. In addition, were we to include dividends, our Liquid portfolio would likely show even higher returns in that some of the holding do pay dividends. 

 

Could you clarify CONCERT AI turnover and performance, particularly outside the market turmoil in 2008?

Turnover varies greatly depending on the market environment. Case in point – in 2012 we had 0 trades; in 2013 we had 34 trades; in 2014 we had 12 trades; in 2015 we had 9 trades; and in 2016 we had 18 trades; and finally, in 2017 we had 0 trades.

Interestingly, during this latest period (May 2016-Dec. 2017) – May, 2016 being the last time the system had any transactions, the stocks selected by the system have returned 43% (not including dividends) while the SPY has gained 29.3%. This is clearly an indication of superior stock selection. Having said this, we clearly believe that our ability to detect the arrival of a less favorable investing environment is a definitive strategic advantage such that when/if a true market correction of some magnitude arrives, our performance should stand out against other programs that are long only. 

 

Could you provide some clarification with regard to ‘curve fitting’ and CONCERT AI's neural networks?

The neural network structure is Mr. Schulenberg’s design, working from first principles and his own ideas, and the training logic was also developed by him. There has never been any sign of 'instability' in these models. He has observed that each neural network seems to have some natural operating 'zones', within which their outputs tend to change fairly uniformly with time, fluctuation, of course, in response to market behavior. There are times when the neural networks tend to generate larger than normal values, and other times smaller than normal values. The magnitude of each value is much less important than its delta changes.

As to over-fitting the data, this is simply not possible; not only have the neural networks been 'frozen' for over 11 years, but they are also very small node-wise -- there are simply too few tunable constants to keep 'fitting' new behavior even if they were actually re-tuned.

There are 11 input nodes in the input layer, one for each independent variable, and each has a pair of associated constants that performs a linear transformation on the data so that all nodes have 'similar' numeric ranges, thus treating oil prices and treasury note interest rates on the same scale, for example. So, these 22 empirical constants are 'scaling' values that ensure that the next layer (the single 'hidden' layer) always sees inputs that vary through the same numeric range.

There is also a 'bias' node in this input layer and it has 12 associated empirical constants.

The main working layer, the 'hidden' layer, only has 9 nodes, thus needing 108 empirically determined constants to connect it to the input/bias layer with its 12 nodes.

Finally, the output layer has 10 associated constants, one for each node in the hidden layer, and a final scaling constant for the output. Thus, there are only 150 numerical values in each neural network, which in turn was trained from 12000 input values -- and these numbers have not changed in well over 11 years.

The key thing here is not the size nor sophistication of the neural networks, but rather the choice of the 11 independent variables that are used to drive them; that is where our proprietary advantage comes in. 

 

What is the difference between "historical" and "pro-forma" results, and how do these measurements apply to your level of confidence in the probable future performance of CONCERT AI?

When creating our charts and statistics for CONCERT AI, we distinguish between historical and pro-forma performance. Historical means that there has been no change or human intervention in that part of the program since the date specified in the summary. Craig refers to these parts of the program as ‘frozen’ meaning that they have been untouched since the specified date. In the case of what we call historical performance, this part of the program was ‘frozen’ in 2004. The stock selection process was ‘frozen’ in 2007. As such, we think it is fair to say that what we call historical performance is in fact forward tested as are other parts of the program.

Having said that, anytime you use neural networks that continue to learn from new data which modifies the inputs to the 69 coefficients that influence the stock selection process, you can’t really say it is forward tested. However, any new data has less influence over time in that the stock selection process uses all history from the creation of the neural networks and their related training period such that any new data represents a smaller and smaller percent of all data used.