Generalized Candlesticks (the RaDiSH Transform™)

Although our short-term signals (Grail System™) are based on our Market Timing (Temperature) Indicator (MTI), and are thus thermally-based, they do not operate on a sufficiently reactive time scale to be able to predict the daily movement of a single equity like QQQ (since the temperature of the stock market can't change very much in a single day). Instead, we turn to an elegant generalization of Japanese Candlesticks. By applying our proprietary transform (RaDiSH Transform™) we can convert the four daily prices (High, Low, Open, Close) plus volume, into a set of 5 dimensionless and time-invariant numbers. Then, by using this data to train a neural network we can predict the average price that a given stock 'may' exhibit on the next trading day (the sum of the High and Low prices divided by 2). Finally, by looking at the actual Open price of QQQ, together with this estimated average price for the next day, we can calculate the probability that QQQ will Close higher than its Open. This gives us a QQQ day-trading signal that is probably about as accurate as is theoretically possible. A sample Prediction Table report can be seen here.

ETF Decomposition of Mutual Funds

Our Mutual Fund Timing Indicator (MFTI) was developed using our Stock Market Timing Indicator (MTI) as a basis, and this works very well to identify those protracted periods when mutual funds (as an investment class) are in either a 'buy' or a 'sell' state. However, the task of developing buy/sell/hold signals for individual mutual funds requires a different approach. Simply looking at the recent price history of a fund shows its current performance, but offers little insight as to how the fund will perform over the next month or two. Our solution is to try to determine the current holdings of the mutual fund, and then to use our intermediate-term buy/sell/hold (and short-term buy/sell) signals to calculate a weighted signal for the entire fund.

Our approach is to decompose each mutual fund into a weighted sum of 5 different ETFs (basis vectors) that are chosen from a large set of 175 ETFs. Monte-Carlo methods are then used to determine the 5 'best' ETFs, and their associated weightings, that reproduce the recent price history of the mutual fund most accurately. These decompositions are often uncannily accurate, and correlation coefficients (R**2) of 0.9900 or better are routinely achieved. While these decompositions can never exactly reflect the current holdings of a mutual fund, they nonetheless give us an analytical way of assigning a net Score (and a buy, sell, or hold recommendation) for over 1,100 funds on a daily basis. A sample mutual fund decomposition report can be seen here.

Differential Signals -- Comparing Two Indices

By employing a mathematical device we can utilize our Grail System™ techniques to generate models for new 'signals' that predict the relative performance of one index against another. We actually utilize ETFs for this purpose, and the objective is to predict whether, for example, QQQ (the Nasdaq 100) will outperform (or underperform) the S&P 500 (SPY), the DOW (DIA), or the Russell 2000 (IWM) over the near-term. A '+/-' value for the $NQ-RU signal, for example, would mean that a simultaneous Long of QQQ and a Short of IWM (i.e., a hedged situation) would likely be profitable; a '-/+' signal would suggest a profitable QQQ Short while going Long on IWM. We currently generate a total of 31 differential signals, and a recent Differential Signals report can be seen here.

Last Changed: 10/19/20