What is the Crest?

The Crest is a trading system based on a computer program. It may be considered as a mutual fund that monitors automatically risks facing investor’s stocks and gives recommendations on what to buy and what to sell. The system trades 500 biggest US stocks and follows the index of S&P500.

When the market is going up the system searches stocks that might be profitable to buy or sell. When the market is going down it sells out stocks from the portfolio to reduce the risk. In that situation it is unprofitable to keep capital in stocks because their value is diminishing. It is better to keep the money on account.

Most mutual funds keep capital in stocks even when market is going down. Very often managers of these funds also fail to monitor market effectively due to psychological reasons like fears, hopes and beliefs. Unlike humans, computer-based systems do not get tired and they don’t have feelings.

The Crest includes 50 different models. Investor can use the most suitable model according to his risk and profit –expectations.

The system is based on latest mathematical methods. Research and development of the system took eight years and cost 500 000 euros. Work was partly funded by National Technology Agency of Finland.

How to use the Crest S&P500?

When investor tests the system, minimum capital needed to use the Crest effectively is 100 000 US dollars. This is because S&P500 companies are big and prices of their stocks are rarely under 5 US dollars/stock.

Due to the nature of US stock market, best profits are usually collected between October and March. Biggest rises in the stock market usually occur during that time. That is why we recommend that investor tests the system for one year, if he doesn’t begin investment activities in autumn. One year is enough to make sure that the Crest works.

If investor after testing period wants to continue to use the system, he will get maximum profit if he uses the system at least three years. This is due to volatility of stock markets.

Year 2004 has been difficult for the US stock market. The market has gone up and down. However, Crest models have produced good profits, 6-18 %.

When the market is going up, investor can expect on average 40 % profit with Crest models. When the market is going down, the profit expectation is +-0 %. This is a pessimistic estimate, because there are models that have produced good profits even when market goes down.

The Crest was tested with real money during the year 1999. It produced 40 % profit. Trading was done by Datek Online Bank and full trade reports are available.

Investment Objectives

• Long term returns in excess of S&P 500
• Lower volatility than the S&P 500
• Low correlation with S&P 500

Business Objectives

• Launching a stand-alone or joint-venture fund.
• The fund can either be (ideally) long-only or “variable long-biased” with a risk overlay (longs model-based, shorts the index). However, it is the view of developers that risk overlay does not improve risk/return characteristics.

1. The Trading System CREST

The trading system is based on a computer program and we call it Crest. Crest acts as an automatic fund manager deciding which stocks to buy, when to sell them and how much of the assets should be retained in cash. The main idea is to invest in stocks that are believed to increase in value during the following four weeks, and if no such stocks are found, stay out of the market. Crest trades only the stocks included in the S&P 500 index. The amount of assets retained in cash increases when Crest detects few investment opportunities and decreases when Crest detects numerous investment opportunities. Crest trades at most once a week. No leverage or short-selling are used.

The track record has been very good since 1999, when Crest was locked up and no modifications have been made since then. We expect Crest to produce annual returns of 40% with 25% volatility during bull markets and during bear markets 20% with 15% volatility. Crest was tested with real money during the year 1999, when it produced returns of 40 %. Trading was done through Datek online brokerage (now part of Ameritrade) and account statements are available.

While we only have one trading system, the system is a collection of overlapping heterogeneous strategies. The collective decision of these strategies is used to initiate long positions and exit existing positions under portfolio constraints. The strategies, collective decision making and portfolio constraints are optimized for best long-term portfolio-level risk-return relationship from investor’s point of view.

Crest is based on latest mathematical methods. Research and development of the system took eight years and cost 500.000 euros. Work was partly funded by National Technology Agency of Finland. All rights are reserved and are the property of the developer. Technical details of the system and description of the development work are available after signing non-disclosure agreements.

2. Investment Philosophy

2.1 Systematic approach

After extensive research we have concluded that the best risk-adjusted returns come from a long-only strategy that stays out of the market when there is a lack of detected investment opportunities. This leads to a market risk in the form of positive correlation between the system portfolio returns and index returns. However, Crest has performed well even during general market declines, as in 2000-2002. Correlation calculated from only negative index periods show that the correlation coefficient, or beta, does not correctly measure market risk of the Crest strategy.

Crest only executes long trades, since shorting is totally different from long-only trading. A market-neutral (i.e. long-short or equity hedged) strategy under systematic approach lowers the expected returns considerably, while volatility is only slightly decreased. Crest minimizes downside market risk, as measured by downside deviation, by being out of the stock market.

Our belief is that the best long-term results come from a number of experienced traders in a portfolio context. Consistency in exploiting correctly the right mispricings can only be attained by using an automated trading system of heterogeneous strategies trained in a machine-learning environment. Including many strategies minimizes the volatility of returns and slows the system efficiency decay. Using computerized systems minimizes expenses, makes the investment process more transparent and lessens human risk, but also introduces new operational risks (IT failure etc).

2.2 Inefficiency is generally misunderstood

This section is a short introduction to the system technology and why Crest is capable of identifying and exploiting mispricings. Financial markets are relatively efficient overall but mispricings occur. Most of the perceived mispricings are the result of data snooping and fail to exist in out-of-sample periods. Often the perceived mispricings are simply risk premia for excess kurtosis and/or skewness, or they ignore significant but less frequent events (peso problem). Therefore the results of using perceived mispricings are often not satisfactory.

Most of the true exploitable mispricings are complex phenomena that cannot easily be analyzed by standard quantitative methods and are normally exploited by experienced traders. While experienced traders can produce good risk-adjusted returns most of the time, there is a possibility that the utilized mispricing-phenomenon is only partially understood, leading to erratic results. Additionally, most mispricing-phenomena will decay in time through markets becoming more efficient.

3. Investment Process

The price- and volume data of the stocks included in the S&P 500 index are inputted to the Crest after the market close at the end of the trading week. During the weekend the system selects the stocks that it deems are about to increase in absolute value. Depending on the number of stocks found and existing portfolio holdings, Crest produces a list of stocks, including the amount, that have to be bought on following Monday at the market open. Crest also produces a list of stocks that have to be sold from the portfolio either because the stock price has risen to target levels, the stock price declined to stop-loss areas or the stock price has done neither after 6 weeks from initial purchase. Fund assets not invested in stocks in any given week are invested in the money market until the next week. We have assumed the money market return to be one week USD LIBOR minus 10 basis points.

Human action is required in only few situations; signal execution and position rebalancing. The signal execution is performed on Monday and cash is invested in the money markets. Rebalancing of positions is required in two cases; during redemptions/new investments to the fund and when a particular stock’s percentage value of total portfolio value changes due to increase / decrease of the stock price. Rebalancing of positions is done at most once a week, and only if the %-value of a stock moves outside predefined boundaries.

4. Risk Control

In our case we view risk to arise from System Risk, System Development Risk, Market Risk, Operational Risk, and FX Risk (depending on the home currency of the fund). System Risk and Operational Risks are the most relevant risks to the successful implementation of the strategy. Of little importance are interest rate, credit and liquidity risks, since the strategy does not use leverage and only invests in short-term money market products.

4.1. System Risk

The System Risk consists of two parts - does Crest continue to be robust, and if not, how can it be detected as early as possible. Crest was trained with historical data, then tested under a previously unseen data set and finally again tested under a new data set. After the final test in 1999 Crest has not been modified and the results have been good since. Also, Crest was tested with real money in 1999 with good results. This is probably everything that can be done to minimize system risk. Since the performance of the system has not changed noticeably after 1999, we believe it will continue to remain robust for at least several years.

By continuous evaluation of Crest’s performance in live trading and careful analysis of the trading performance we believe that loss of robustness can be detected. Parameters for this analysis include the number of detected buy- and sell signals in sub-periods, their subsequent absolute and relative returns, risk-reward ratios and probability of ruin calculated from the signal returns.

4.2. System Development Risk

The System Development Risk arises directly from System Risk. We assume at least some of the inefficiencies exploited by the system are transitory, but the half-life of each inefficiency varies. More importantly, we believe that new inefficiencies will be born in the future. This means that periodical overhaul (between 3-4 years) of the system could improve system’s risk-return characteristics and lessen the correlation to index. Further development has to be balanced with a) the cost of R&D, and b) dangers of data snooping and c) danger of abandoning a well-functioning system for a new system that is unproven. By the academic term data snooping we refer to a situation when a spurious correlation detected in recent data is misinterpreted as causality. With limited data sets this becomes an issue, since a detected anomaly or inefficiency in recent data cannot be verified with previously unseen and new data.  Abandoning a well-functioning system basically means that low temporary performance is misinterpreted as permanent performance loss, and a switch to a new and “improved” system is made. This creates two sources of trouble – the old system could still be robust, and a future positive run is missed. The new and “improved” system could fail to perform out-of-sample. The System Development Risk is believed by us to be minimal for the next several years. System Risk and System Development Risk are closely linked.

4.3. Market Risk

Market Risk is the only readily measurable and exogenous variable for the system. The market risk can arise from market shocks (i.e. stock market crash), naïve diversification leading to strong correlation between individual positions, and liquidity issues affecting market impacts and bid-ask spreads. While long declining market conditions do obviously hurt the profitability of the system, the performance during the 2000-02 bear market was very reasonable. The worst scenario is a market meltdown, which would cause crash in all stock prices, fall of liquidity and increase in bid-ask spreads.

4.3.1. Hedging Systemic Market Risk Not Reasonable

While the correlation of trading system returns and S&P 500 index returns do show a moderate level of positive correlation, the correlation arises mainly from positive return periods. Correlation, when calculated only from periods when the S&P 500 returns were negative, is small.

A simple hedging strategy of the systemic market risk has been tested, where a hedge with S&P 500 futures is established in the size of the current investment, i.e. if the current investment ratio is 50% and total assets are 10 million USD, a short position of 5 million USD is established in S&P 500 futures. While this lowers the correlation between trading system returns and index returns to effectively zero, the system return volatility is decreased by only couple of percentage while the returns are considerably smaller.

On a risk-adjusted basis hedging the systemic market risk does not therefore make sense. Of course, potential investors might be more attracted to a fund with a market-neutral strategy, but the effect of a more attractive style might not be enough to overcome the lower risk-adjusted and absolute returns and lower performance fee income. Introducing hedging of systemic risk also introduces costs, operational and liquidity risks.

4.4. Operational Risk

Management of operational risk is critical for the successful implementation of the investment strategy. It can be subdivided to several subcategories:

• IT failure
• System signal disobedience
• Fire etc.
• Theft & Fraud

4.4.1. IT failure

IT failure can be either catastrophic or creeping. In catastrophic failure the input data, actual system, trading orders placed through Internet etc. fail completely. This can be minimized by redundancy - multiple locations and equipment, several data vendors (with data checked manually for obvious errors and differences), several Internet and telephone lines. Creeping IT failure is hard to detect – it can be a small programming error in the actual system, failure of processors, operating systems or software applications to function in the intended way. Cross-checking, redundancy and following industry media are the best remedies.

4.4.2. System Signal Disobedience

System Signal Disobedience (SSD) means that the trading signals are not executed as intended. Signal can be completely ignored, trade execution can be delayed or the trade is only completed in partial amount. The fund trader second-guessing either the system (the system will sell this anyway next week) or the market (I can buy this stock at a cheaper price tomorrow) are seen as the most important reason for SSD. Therefore a contractual obligation to execute the system signals as provided have to be signed by the relevant personnel.

4.4.3. Fire, theft, fraud  etc.

Loss of key equipment, communications or documentation for any reason can be remedied with redundancy using multiple locations, communication lines, and weekly system back-ups. Also proper physical security measures are called for.

5. Questions & Answers

5.1. Investment Strategy

Do you employ a single investment strategy or multi-strategy?

While there is only one trading system, the system is a collection of overlapping heterogeneous strategies. The collective decision of these strategies is used to initiate long positions and exit existing positions. The strategies and their collective decision making are optimized for best long-term portfolio-level risk-return relationship.

Are the market inefficiencies you exploit present continuously or do they appear sporadically? What market environments favor or hinder the availability of investment opportunities?

The market inefficiencies we exploit are constantly present in the markets, but their magnitude and number varies. There are relatively long time periods when the system does not make any new investments or even stays completely out of the stock market. The absolute performance of the system is positively correlated with the S&P 500 index, while the relative return and risk/return relationship are superior to S&P 500 index.

How are investment ideas generated and how are they implemented? What are your primary sources of research? What securities and instruments are used to implement your views?

The investment idea generation is fully automated. Only the stocks included in the S&P 500 index are used to implement the investment ideas.

Do you use external investment sub-advisors? If you do, identify them, describe their contribution to your investment process, and describe how their performance and compliance with your investment guidelines are monitored?

No external sub-advisors are used.

Describe your criteria for a new investment. Are your investment criteria purely objective (i.e. all investments that meet a fixed set of criteria are entered into or are they partially or wholly subjective (i.e. investment criteria are not entirely predetermined and/or each investment is considered on its own merits possibly using its own unique set of criteria)

Investment criteria are purely objective.

Are new investments considered in isolation (i.e. purely as a good long holding) or as part of a hedged basket (e.g. as one leg of a pair)

The system does consider new investments in isolation, but under portfolio constraints.

What is the expected return of the portfolio? How do you arrive at this number?

Expected return of the portfolio is 40% during a year of generally rising stock prices. The number is based on historical simulations before 1999 and after that on live simulations.

How and when are existing positions in the portfolio closed out?

Existing positions are closed when system decides so. This happens at the earliest opportunity without human discretion. The exit decision is based on several factors, including, but not limited to: trailing profit, take profit, pivot points, stop loss, trading range and other technical signals and pattern recognition signals.

What level of turnover do you expect? What is your average holding period? Investment ratio?

The average annual turnover is expected to be 4-5 times. The average holding period for a single stock position is four weeks, with a maximum period of 6 weeks. The investment ratio (assets invested in stocks) varies between 0 and 100%, with an average of 35-40%.

What fraction of a day’s trading volume do you limit your positions and trades to?

Initial position in a stock is limited to 10% % of total portfolio assets. After the initial trade this value may increase or decrease when the stock price increases or decreases, or alternatively because the value of other stocks in position changes. The average daily volume of a stock traded by the system has been slightly over 100 million USD. A fund with investable assets of 20 million USD would therefore have an average position size of 2% of daily volume. Currently no limits have been set.

What is the capacity of your strategy, and what is the primary determinant of this? At what level of assets will the fund be closed to new investors? At what level of assets will the fund be closed to existing investors?

The capacity of the system strategy is currently undefined, but large. Primary determinant is the market impact of trades. Since the system trades only once a week, the strategy is not as susceptible to market impact effects as higher-frequency strategies. We estimate that with properly executed position building the fund could handle 100 million USD, at which point single position would be 10% of average daily volume. The capacity can be increased by increasing the number of positions from the current ten, but this would have a negative impact on the returns. We plan to study the market impact question deeper in the future.

Do you hold any illiquid securities such as micro-cap stocks?

No. At an early stage in the development work the whole U.S. stock universe was studied, but current system trades only S&P 500-stocks.

What is the minimum, average, and maximum exposure to cash? How is cash managed?

There is no predefined maximum exposure to cash. Minimum exposure to cash is set at zero, since no leverage is employed. Average exposure, based on historical averages, has been 30-40% and this is believed to be the future average as well. System simulations assume that cash is invested every Monday at USD one week LIBOR rate minus 10 basis points.

What are your borrowing costs and what do you earn from your short positions?

The strategy does not employ leverage and does not therefore have borrowing costs. Also no short selling is allowed.

5.2. Risk Control and Portfolio Construction

Do you calculate VaR and do you Stress Test the portfolio?

We believe that VaR analysis and Stress Testing are not valid methodologies (see e.g. Nassim Taleb’s criticism at www.fooledbyrandomness.com). However, they are important because of lack of better methodologies and regulatory purposes. Calculating VaR numbers or stress testing can easily be done under our system, and we can provide the numbers on an ad hoc or regular basis.

Do you have risk limits for market risk?

The risk limits for market risk are automatically set by the system, so that the investment ratio varies between 0 and 100%. This is not the same as market risk measured by beta, however.

How do you view diversification and how is your view reflected in your portfolio?

Diversification is the only free lunch in the financial markets and is very important to us. Diversification can be used in many dimensions, including number of instruments in portfolio, time and position scaling. However, diversification does have its limits. Increasing the number of instruments in the portfolio to very large does not necessarily improve risk-return characteristics except in a near-random portfolio management (i.e. most funds).

Our view of diversification is reflected in the portfolio as a relatively moderate number of stocks in the portfolio to provide reasonable diversification benefits (to lower return volatility) but not to increase the correlation to index.

What is the minimum number of long and short positions in the portfolio? How many long and short positions are currently in the portfolio?

A maximum of ten positions in different stocks are allowed under the system constraints. No minimum is set. No short positions are allowed.

How are hedges constructed? Does each long position have a corresponding short position or do you view your portfolio as a long basket offset by a short basket?

No hedging is currently used. Since the system only trades U.S.-listed stocks nominated in U.S. dollars, hedging the currency risk could easily be done on a weekly basis and adjusting the hedge in case of significant variations in position value during the week. The market risk can just as easily be hedged through index futures, but we do not recommend this, since the system is already optimized to provide best possible risk-adjusted returns irrespective of market environment. While hedging the systemic risk through index futures lowers the correlation between system returns and index returns, volatility of the system returns is decreased only slightly and returns decrease considerably.

tial position in a stock is limited to 10% % of total portfolio assets. After the initial trade this value may increase or decrease when the stock price increases or decreases, or alternatively because the value of other stocks in position changes.

What is the range of permissible gross exposure to a single industry or stock?

No industry limits are set by the system, although their effects have been studied during the development. Group behavior of stocks of a particular industry had no positive effect on system performance characteristics.

Is the portfolio beta neutral in each sector? If not, is there an allowable range of sector betas?

No. There is no allowed range set for sector betas.

What level of leverage (min/max/avg) do you employ? How do you define leverage? Does it include off-balance sheet items?

We do not employ leverage. Leverage is defined as something not bought with cash. No off-balance sheet items exist.

What is the volatility, measured by annualized standard deviation, of the portfolio?

The volatility of the portfolio, calculated from weekly observations, has been 20%.

Approximately, how long does it take to sell all portfolio securities for redemption request?

We assume that during at least remotely normal market situations the whole portfolio can be liquidated during the course of one day. The only real difficulties would happen during general market meltdowns (1987 revisited).

Maximum drawdown?

5.3. Technology

1. Does the investment process require computer technology to allow its implementation or can it be performed manually in large part?

The system requires computer technology. The system rules can be exported in a text format, so that the results can be confirmed by manual calculation. The system has been trained under a parallel processing cluster, but the day-to-day system running can be done on a regular PC.

2. How often do you backup your computer systems? Are backup tapes maintained off-site? Do you have a formal disaster recovery plan?

Relevant at the point of setting up the fund. The developer is an IT professional and understands the importance of this. The system includes a shadow-function, which continually records and duplicates all system events to a clone system. In case of system crash, the system automatically switches to the cloned systems. In fatal crashes cloning allows the system to be raised up in few minutes.

The cloning is based on Intersystems-technology, which is used by e.g. Shell, Chase, Ameritrade and U.S. Defense Department in mission-critical computer systems. The largest PC network in the world, including over 40,000 workstations, is based on the technology. The system has been extensively tested by U.S. Defense Department’s  Department of Advanced Research Projects, DARPA.

3. Are your computer systems maintained by an in-house staff or is this function outsourced?

In-house staff.