Alphability: Who will be able to see what?Anyone: can see this page outlining the concept of Alphability and will also have access to a limited range of statistics. Registered Users: will also have access to additional more detailed Alphability statistics. Subscribers: Will be able to see the full range of metrics including detailed breakdowns and will also be able to download historical Alphability data. 
To download this article in PDF format, please click here
An Introduction to Alphability
Alphability statistics are intended to provide an indication of the relative ease of capturing alpha (excess returns) from specific markets and time frames using generic categories of trading model. The two model categories initially supported on www.automatedtrader.net are trend and reversal. (Metrics and a database of matching engine rules for scalping and other ‘order book’ type strategies will be added later.)
Alphability statistics have two main objectives:
To provide historical measures that will assist those building trading models/systems in determining which markets and timeframes are most favourably responsive to which generic types of model.
To provide a hypothetical benchmark for measuring the alpha capture efficiency of trading models of certain generic types. (Not all Alphability metrics are suitable for this second objective).
The second of these objectives is not an entirely new idea; the concept of ‘potential profit’ was outlined in 1992 by Robert Pardo and has since been extended by Valerii Salov in his recent book “Modeling Maximum Trading Profits with C++”. However, in the case of Alphability, the idea is taken a step in a new direction in that it calculates potential profit for specific categories of trading model in a variety of time frames.
Example
Volume Quartile Trends are one example of Trend Alphability and are calculated as follows. The trading volume for a trading session is divided into four equal volume segments that are used to delineate four trading periods. The first period lasts from the market open until 25% of the day’s total volume has traded; the second from then until 50% of the day’s total volume has traded; and so on. (Where actual traded volume data are not available, tick volume is used.)
Figure 1: Individual and total Volume Quartile Trend Alphability metrics 
The ranges from the highest to the lowest prices in each period are then calculated to arrive at the maximum potential return for each volume based period. Figure 1 below illustrates these individual periods as well as their daily totals for 20 trading sessions of the Sep 07 DAX stock index future. All values are in points
This provides four hypothetical ‘potential profits’ that can be used as benchmark for low frequency intraday trend following models. However, this does not give any indication of the ease or difficulty of capturing those four potential returns.
One way of addressing this is to calculate the R2 (coefficient of determination) of a linear regression line from the low to the high of each volume period. The higher this value, the better the fit of the data points to the line and by implication the easier it should be to capture the trend (subject to a particular caveat mentioned later). Figure 2 illustrates the R2 values derived from the closing price of 5 minute bars for the same period as Figure 1.
As can be seen from the Q3 values highlighted on the columns marked ‘A’ and ‘B’, there are considerable discrepancies among R2 values. The reasons for this become apparent when one examines the data points used in the linear regression calculations for these values. Figure 3 shows the data points for column ‘A’, Figure 4 those for column ‘B’. In Figure 3 the progression of the market from the first to the last data point hardly characterises a smooth and orderly trend. While Figure 4 is by no means perfect, the trending nature of the market is immediately evident.

Figure 2: Individual and total Volume Quartile Trend Alphability R^{2} value 
However, R2 values alone can be misleading as they say nothing about the length of the potential trend. A violent four bar move of 5% might have a high R2 value but would be difficult to capture. By contrast a gradual and steady rise or fall over 30 bars might have a lower R2 but be considerably easier to capture. Therefore one variant of the downloadable Alphability metrics available shortly to AT subscribers includes weightings for both R2 and length of move. In addition, subscribers will also be able to download additional data on Alphability distributions, historical datasets and R2 calculations based on high/low values.
Figure 3: Data points for column 'A' in Fig. 2 
Figure 4: Data points for column 'B' in Fig. 
Conclusion
Alphability metrics aren’t silver shrapnel and they aren’t trading models. However, the intention is that they may facilitate the more rapid development of automated and algorithmic trading models. According to our readers, the half life of alpha capture models in particular continues to decline, so the pressure to up the trading model development work rate is definitely increasing. Alphability will never replace rigorous and thorough model testing and evaluation but it will hopefully assist traders and developers to zero in more rapidly on markets and/or timeframes best suited to particular types of trading model. The other goal is that Alphability will also provide a range of granular and appropriate benchmarks for particular strategy types.
While some readers have already given us their valuable input, we see Alphability as an evolving project that must ultimately be driven by practitioners if it is to be of practical value. So if you have any specific suggestions regarding instrument coverage, time frames or other enhancements, please contact Alphability's creator, Andy Webb.
 Prepare failed: (144) Table './ATStatistics/logger_ads_imp' is marked as crashed and last (automatic?) repair failed