Competitors

It has been widely accepted by the investment community the best way to maximize returns across every level of risk is to combine uncorrelated assets, or more recently, uncorrelated passive investment vehicles designed to mirror asset classes. Asset classes vary greatly across competing platforms. Each generally identifies a set of asset classes based on each proposed asset class's historic behavior under different economic scenarios, risk-return relationships conceptualized in asset pricing theories, and their expected behavior going forward based on trends and the macroeconomic environment. In addition, they typically evaluate each asset class on its perceived potential for capital growth and income generation, volatility, correlation with other asset classes, inflation protection, cost to implement via selected investment vehicles and tax efficiency. Much of the analysis is subjective and subject to error. Some vendors have identified as many as 20 asset classes extracted from three broad categories: stocks, bonds, and inflation assets. For example, Growth Stocks could be segmented into: Large Cap Growth Stocks, Mid Cap Growth Stocks, Small Cap Growth Stocks, Emerging Market Stocks, just to name a few.

folioPilot

folioPilot takes a different approach. It starts with the broad investment vehicle classes: Bonds, Stocks, ETFs, Index Funds, Mutual Funds, and Money Market Instruments. Next, it uses pre-defined, pre-validated smart-beta, quantitative investment strategies to segment each investment vehicle class into the more common asset classes used by other platforms. For example, within the asset class "Stocks" folioPilot includes investment strategies designed to limit investment to the asset class "Growth Stocks", another to "Growth Large Cap Stocks", another to "Consumer Cyclical Growth Stocks", another to "Dividend Growth Stocks", and so on. folioPilot does the same for "ETFs". It includes ETF investment strategies designed to mirror not only the system's stock-based asset classes, but many of the more common indexes. (folioPilot's technology also supports Direct Indexing should that be of interest to the advisor or its client.) Through this approach folioPilot is able to build solutions utilizing anywhere from one to twenty-five investment vehicle & asset classes where each class is defined by pre-validated strategies. Generally, adding more classes improves the risk-return tradeoff, but there is a limit. It gets increasingly difficult to improve a portfolio already diversified across seven or eight classes (i.e., in our case, strategies).

Competitors

Most of our robo competitors use low cost, index-based Exchange Traded Funds (ETFs) to represent each asset class. A very few use actively managed mutual funds, but their costs are higher than those associated with ETFs and it's been recently shown that few actively managed mutual funds manage to outperform the market. As a result, index funds and passive ETFs have exploded in number over the past 20 years. However, unlike mutual funds and Index Funds ETFs do not have a standard rating agency, which makes it difficult for the average investor to understand their costs and determine which are best to represent each asset class. Our competitors, through a variety of subjective methods, attempt to choose specific ETFs (or Index Funds) that are ideal in the context of the overall portfolio after-tax risk-adjusted return, but that introduces a selection bias into the portfolio, which is generally not a good thing.

folioPilot

folioPilot does not rely solely on passive ETF investment vehicles. Rather, it uses individual securities drawn from multiple investment vehicle classes to achieve the objectives of Modern Portfolio Theory (MPT). In addition, it does not utilize subjective judgement to pick its investment vehicles; it relies on the rules of each of its pre-validated strategies to pick the appropriate investment vehicles for each investment vehicle class. The process starts with the application of the strategy's buy rules to create a universe of qualifying securities. The candidate pool is then ranked from goodness to badness in accordance with a set of ranking rules again defined by the strategy. Next, the highest ranked securities that meet the buy rules are placed in the portfolio subject to a set of sell rules. It's a rules-based approach where subjective judgement and emotion play no part.

folioPilot's validation engine is a core technology and is unique to folioPilot. It's built upon proprietary technology adopted from supercomputer aerospace simulation technologies with which the Company's founder is familiar. More specifically, the Company has based its simulation techniques on techniques first developed by NASA to simulate the flight of advanced, high performance aircraft where each pressure point surrounding the aircraft is pre-computed. Like the pre-computed pressure points, folioPilot's strategy environment is pre-computed and searchable. This enables the search for strategies that meet specific client requirements in terms of ROI, risk, and investment horizon. In addition, the engine has been highly optimized to take advantage of parallel processing algorithms and other little known features included in today's hardware platforms, a prerequisite for overnight updating and rebalancing.

folioPilot's database of historical securities data is central to the engine. It includes both fundamental and technical data on active securities as well as inactive securities. As a result, the database is without survivorship or look-ahead biases; it is a "point-in-time" database. When the simulator tests an investment strategy for its historical trading behavior and performance, it simulates it daily over as many years as 30 utilizing only the data known at the time of the simulation. To do that, the database contains only "as reported" data, that is, it does not contain corporate restatements until such time as they were actually published and made available to investors.

Competitors

All managers, robo or otherwise, collect sufficient information about their client to meet the standard of care required of them in relation to the scope of services agreed to by the client. The appropriate question is therefore not how much or type of information a manager collects, but rather whether the information the manager decides to collect is appropriate in relation to the nature of the services provided. It follows then that our competitors need not collect the same information or conduct comparable due diligence, to that which we have to collect for a more expansive service. Point is, we all collect KYC information, but we do it differently and we interpret it differently making a comparison of methods difficult.

folioPilot

The aforementioned notwithstanding, folioPilot's onboarding process has been designed to identify a client's financial goals, investment horizon, and risk tolerance in order to select from our library of strategies the ideal set of strategies and client allocations to those strategies needed to meet client objectives and needs. KYC information for institutional portfolio owners is out of necessity different in nature.

Competitors

The vast majority of our competitors determine the optimal mix of asset classes by solving the “Efficient Frontier” using Mean-Variance Optimization, the foundation of Modern Portfolio Theory (MPT). The approach was popularized by Nobel Prize winner Harry Markowitz back in 1952 making it not really very modern. The Efficient Frontier (a chart) represents the portfolios that generate the maximum return for every level of risk. Each portfolio is created by choosing a particular mix of asset classes that maximizes the expected return for a specific level of risk (as measured by variance), or equivalently minimizes the risk for a specified expected return. Mean Variance Optimization (MVO) calculates the best risk-return tradeoff when combining the asset classes into portfolios. The goal is to find the portfolio with the asset allocation mix that is the most efficient. In other words, of all of the investments fed into the optimizer engine, it’s the mix that has had the most return per unit of risk over the selected time frame that wins the day.

folioPilot

In creating folioPilot, we've taken a very different approach. So, why did we pass on Mean-Variance Optimization? It's because asset class correlation coefficient numbers are too random. Two asset classes that have moved in opposite directions over the last ten years could have moved in sync with each other over the last five years, or just one-year, or last week. And more importantly, in a major down market, they always move in sync with the market, always down, just when diversification is most needed.

Optimizing at the investment vehicle level is also problematic as their returns, and the resulting correlation coefficients are event driven - so they change daily based on millions of people making investment trades based on how they react to the daily news. This means they can't be predicted, they change as the world economy changes, and they change dramatically when you change the benchmark slightly.

The bottom-line is that one can't really use correlation coefficients to forecast what's going to happen over any future time horizon. This is because they are long-term averages, which mean an investor will most always invest right when there is the highest correlation, and then sell right before the low point in correlations.

So, how does folioPilot optimize a client's portfolio? It uses non-linear optimization which does not require correlation numbers. The only two numbers needed are expected return and expected standard deviation, both of which it gets from its strategy simulations. During simulations folioPilot calculates for each investment horizon a strategy's expected return as well as its expected risk or standard deviation. From the client's profile folioPilot ascertains their desired ROI, risk tolerance, and their investment horizon, which enables folioPilot to calculate an optimum allocation to each class specific strategy. Should a client's investment horizon, desired ROI, or risk tolerance change, folioPilot recalculates the optimum allocation of capital between the system's recommended strategies which then leads to a portfolio rebalance.

In addition to calculating optimum allocations of capital among classes and the strategies used to define them, folioPilot enforces minimum and maximum allocation constraints for each class or strategy. This method is widely used to ensure proper portfolio diversification, mitigate parameter estimation errors and express client preferences. For retail portfolio owners, folioPilot uses 5% as a minimum allocation because anything less than that does not provide meaningful diversification. In addition, it uses 40% as the maximum allocation to ensure sufficient diversification from meaningful allocations to the other classes or strategies.

Some portfolio owner's may specify a single class portfolio, for example, an ETF Wrap, stock, or fixed-income only portfolio. In those situations, folioPilot allocates capital equally among the strategy's recommended holdings. Why? Over the past 36 years we’ve found this approach to be superior to Mean-Variance Optimization for the reasons set forth above.

Competitors

Since a portfolio created using MPT-based techniques will not stay optimized over time, they must be periodically rebalanced in order to maintain the intended risk level and asset allocations. Competitors employ a range of methods (time based: bi-weekly, monthly, quarterly, semi-annually; or threshold based: when dividends accrue, a deposit or withdrawal has been made, or if movements in their relative allocations justify a change). Some competitors rebalance subject to tax and trading expense effects.

folioPilot

Since folioPilot's portfolio optimizing methodology differs from that of our competitors, the need to rebalance client portfolios is less often, which works to the client's advantage. On a daily basis, folioPilot monitors client portfolios relative to the rules of the strategies chosen for each investment vehicle class initiating a rebalance when a holding no longer meets a strategy's rules or when the portfolio just needs to be re-optimized in order to maintain its target allocations among the chosen class-specific strategies. Like our competitors, folioPilot initiates a rebalance when significant dividends accrue or a significant deposit or withdrawal has been made. What constitutes a significant event is user adjustable.

Validation is one of the most important elements of building an investment strategy. Unfortunately some investors have come to devalue validation (simulation) primarily due to poor correlations between their results and a future that doesn't turn out to resemble their test results. As we discuss here there are many reasons for these negative outcomes, but we remain steadfast in our belief that thoughtful validation is a must in order to distinguish between ideas we think ought to work versus those that have a reasonable probability of working.

Survivorship Bias. One major problem we often see in competitive back-testing is known as "survivorship bias." This refers to tests conducted using only securities currently traded. If that's the case, securities no longer in the database would never have been considered in the analysis. This is the survivorship-bias problem – the test results are said to be biased because they reflect only those securities trading as of the date of the test. The results would exclude all matured or delisted securities. Sometimes, survivorship bias will make the results look better than they should; i.e. the test would not take into account securities whose values dropped as a result of poor performance or a corporate action. Other times, survivorship bias will make results look worse than they should; i.e. the test would not take into account securities whose values rose as a result of a corporate action, rollover or redemption. There's no way to know which way a particular test's results will have been biased. All we know is that the bias exists and it can be substantial.

To address the issue folioPilot's database is a point-in-time database that includes all securities including those no longer listed on an exchange. When running a validation, folioPilot includes securities up until the time when they cease to trade. This means that if you run a validation today delisted securities will be included in the results whenever they passed a model's criteria and for as long as the securities remained a part of the portfolio and continued to trade.

Additionally, it's not wise to create a model using tickers versus names or security identification numbers. Why? Because data vendors replace tickers of delisted securities with new tickers that flag those securities as delisted. This is essential because delisted tickers are often reassigned by the exchanges, even tickers that were attached to very high profile securities. Failure to not give this issue its due during the construction of a test will result in highly skewed results not reflective of the intent of the test.

The fact that folioPilot uses a point-in-time database is vital.

Classification. Not only must the testing platform avoid survivorship bias, it must also take into account security classifications. For example, business classifications. Here's a representative case of how a S&P GICS classification can change over time. Presently they classify IBM as being in the IT Services Industry and the Information Technology Sector. Before 2010, it was in a different industry within the same sector: Computers & Peripherals. This is important. A back-test could be distorted if the database were to use only the current classifications for all points in time. Historical correctness is necessary because models being tested may screen on industry averages or ranking factors that are sorted relative to industry and/or sector peers. A testing platform that evaluates IBM as an IT services company in a 2000 instance of the model could produce erroneous results.

Stock Splits, Stock Dividends and Cash Dividends. It's interesting that so many modelers fail to consider these topics given their importance to proper testing. folioPilot's validation results are computed on the basis of data that has, where appropriate, been adjusted for the impact of stock splits and dividends. The split and stock-dividend adjustments are present in pre-calculated ratios that depend on the number of outstanding shares, such as EPS and also in the ratios involving shares calculated using custom functions. Again, these items can have a significant impact on computed results and should not be overlooked.

Restatements or Reclassifications. It would be easier if security issuers could simply report their results after the end of a period and move on. But things are often more complicated. For example, it's not uncommon for companies to revise previously-reported results. Whether and under what circumstances restated data should be used in a test is not as clear-cut as one might suppose. We take the position that as of a point-in-time, a test ought to include only that data that would have been available to the investor on that date and not on the basis of changes investors couldn't possibly know would come about in the future. So when we're concerned about the relationship between company fundamentals and share price trends, historical stock prices need to be seen together with the data available to investors at the time. Too often inexperienced modelers fail to account for restatements or reclassifications thereby further impacting computed results.

Look-Ahead Bias. Survivorship bias isn't the only kind of bias against which we must guard: There's also look-ahead bias. The mission of a point-in-time database obviously involves the recording and cataloging of accurate data, but that's not all. It's important to be able to discern when data items became known to investors. The availability of information and when-did-we-know-it is often easy. Either the information has been disclosed and we know it, or it has not yet been disclosed and we don't know it.

Trading Strategy Back-Testing. So by now it should be clear that the above challenges could and often do impact testing results. But they're not the only factors that influence investor perceptions of back-testing. For example many of our competitors offer some form of back-testing more often than not limited to testing and comparing the viability of a trading strategy. The goal is to test and compare various trading techniques without risking capital. The theory is that if a strategy performed poorly in the past, it is unlikely to perform well in the future (and vice versa). The two main components looked at during testing are the overall profitability and the risk level taken.

Implementing a trading strategy back-test usually involves running a model (i.e., strategy) via simulation. The simulation is run using historical intraday pricing and/or volume data as inputs into any number of technical indicators (e.g., Moving Averages, Bollinger Bands, Relative Strength Index, etc.) The goal is to identify and execute "buy" and/or "sell" signals. It's also essential that the model be tested across many different market conditions to assess performance objectively. Variables within the model are then tweaked for optimization against several different back-testing measures.

Investment Strategy Validation. folioPilot's approach to testing differs in a number of important ways. But first it's important to understand the difference between "traders" and "investors". The main difference is the duration for which a person holds the asset. Investors have a longer-term time horizon, whereas traders hold assets for shorter periods of time (e.g., minutes, hours, days) to capitalize on short-term price movements.

Due to the difference in time horizon, trading strategy back-tests use intraday or "tic" data, and only for the security or portfolio specified in the model and only for the time period over which the model is being run. The enabling database typically consists of just a few fields, but potentially a lot of time periods resulting in terabyte size databases.

The testing of a folioPilot investment strategy requires the use of interday data (e.g., open, high, low, average and closing prices, volume, and most importantly fundamental data drawn from SEC filings). Examples of fundamental data include revenues, operating expense, profitability, earnings per share, cash flows, leverage ratios, etc. Since a security's fundamental data does not change throughout the day, intraday data is not needed to test an investment strategy. Further, our asset class datasets include a minimum of 30 years of daily data (a time period long enough to include major market movements) and up to 2,000 fundamental and Technical fields for each security. As a result, our database contains thousands of fields but fewer time periods than a typical database built for a trading back-test.

Finally, since our database structure and its elements differ from that required for a trading back-testing system, we use an alternative form of simulation. But first a couple of definitions; a computer model is the algorithms and equations used to capture the behavior of the system being modeled. By contrast, computer simulation is the actual running of the program that contains these equations or algorithms. Simulation, therefore, is the process of running a model.

folioPilot's Strategy Builder enables the construction of a multi-variant model that contains a set of buy rules, a set of ranking rules, and a set of sell rules each set forth as an equation. Taken together they represent a strategy or "model". folioPilot has taken a "continuous dynamic simulation" approach to running its models. This form of simulation solves each equation and then periodically, solves all the equations and uses the numbers to change the composition of the portfolio's state and output of the simulation. To be clear, a folioPilot test is not designed to forecast future performance, it is designed to give the user confidence in their investment approach. Should the user decide to fund their approach, the system will then monitor their portfolio daily in order to alert the user to those positions that no longer meet the rules of their strategy and suggest other positions which do.

Trading model developers typically use discrete-event simulation to run their models. Their approach runs a model as a (discrete) sequence of events in time; each tic being an event. Each event occurs at a particular instant in time and marks a change of state in the portfolio due to changed input signals, e.g., prices. It's a much simpler approach as it does not recommend positions or monitor positions for non-compliance.

Summary. Experiences with testing vary widely. It's not only the challenges described above that impact the testing experience, but the accuracy and latency of the data, the appropriateness of the simulation method used, and finally the skill set of the modeler. All play a part in creating the experience. We've been at this since 1981. We're considered to be among the best of the best at what we do.