The Gateway to Algorithmic and Automated Trading

Going Green

Published in Automated Trader Magazine Issue 12 Q1 2009

If you are looking for an order execution edge, the solution may already be sitting in your dustbin. Shaun Downey explains the practicalities of recycling your stale alpha models for algorithmic trading.

One opportunity still underexploited in the world of trade execution is the use of technical analysis for algorithmic order placement. This opportunity can take at least two forms:

• Purpose-built algorithmic execution models using technical tools, typically in timeframes below ten minutes
• The re-use of existing technical analysis based trading systems or filters (alpha capture models)

In many cases, the latter approach allows an element of intellectual property recycling. For example, an alpha model with decaying performance might once have an average trade profit of twenty pips but over time this has declined to perhaps three pips. As an alpha capture model it might therefore be considered junk. But what if it can be redeployed as part of the execution process for another newer alpha model? While three versus twenty pips as the average alpha capture sounds uninspiring, three pips price improvement on every trade execution by another model is rather more appealing…

The snag with this recycling approach is that many such stale alpha models trigger price entries on the close of the price bar. In the case of popular timeframes, there can be significant fluctuations in execution risk with this scenario. This can be
especially pronounced when common 'crowd following events' (such as stochastic crossovers) happen to coincide.

Recent analysis by Chiron Investments on a variety of futures markets (conducted in MATLAB using tick data supplied via CQG's API) revealed that the minutes immediately prior to price bar completion in higher time frames (e.g. thirty or sixty minutes) typically display lower range than the minutes immediately after bar completion. Figure 1 illustrates this difference for thirty minute price bars of the Dax future calculated in points from a year of intraday data. It also highlights a similar diversity pre and post a popular technical signal, namely a crossover in the slow stochastic.

Analysis of volume shows that it is also higher after the bar's close, but that this rise is insufficient to compensate for additional potential slippage caused by the expansion of range in the critical one to three minutes after bar completion. This makes algorithmic trade execution immediately post bar closure more demanding, as the ratio of liquidity to range is lower than immediately before.

For automated models, where pre-empting bar closure is not possible, this has obvious implications for performance in terms of slippage. However, technical tools are available that can be used to address this at the micro level. A case in point is the Range Deviation Pivot, which has an in-built skew to account for any trend. Once a bar opens, the study calculates and places +/- one, two, and three, standard deviation levels around that value. If the trend is down, then
the pivots above will be closer to the opening price than the pivots below, allowing room for the trend but narrowing the risk for a reversal. A simple but effective method for using this tool is to look for a close beyond +/- one standard deviation on a one minute bar as a reversal signal for the next bar.

Figure 1: Average of range three minutes pre- and post- closure of Dax thirty minute price bars

Figure 1: Average of range three minutes pre- and post- closure of Dax thirty minute price bars

Figure 2 illustrates the performance of this tool in the one minute time frame, as calculated by CQG's Entry Signal Evaluator. When the signal is triggered, subsequently delaying entry by anything from one to three minutes shows an execution improvement of one to four ticks. (The test was run on a small portfolio of commodity futures.)

Whilst the example here is for one minute bars, extensive testing has shown the benefit of using similar relationships on daily data as a method of pyramiding trend following systems. This is achieved by exploiting intraday reactions to the dominant trend instead of the more common practice of pyramiding based on the closing price. Put simply, if the trend is down, closes are more likely to be near the lows of the day, therefore increasing instability in the system's performance.

Range Deviation Pivots by contrast typically increase the overall profitability of an existing model but not at the expense of stability. In the case of position management, a typical example of their usage would be to take an existing short trade in a downtrend and have a limit order to add to the short position placed at the plus one standard deviation pivot. (The pivots can also be used as a 'disaster' or 'trend ending' stop if price closes beyond the third standard deviation.)

Volatility Time Bands (see Figure 3) are an alternative tool that can be used in higher timeframes to provide an edge to order execution, but can also be viable as a pyramid point or qualifier for an existing trade. These bands track momentum by referencing the specific time of day of execution and then computing the normalised range for that time of day. This creates a concertina effect, as the bands ebb and flow through the day. Once again, the values are based on the opening price of the bar and therefore provide a fixed reference point of symmetrical distributions during the entire life of the bar.

There are a multitude of patterns and relationships that can be exploited with this tool. The study can be used to finesse algorithmic execution but also to gain a better understanding of execution risks depending on the time of day a trade is being placed. These can vary enormously, especially since the trading signal outputs of most technical or quantitatively based alpha models are random in terms of the time of day they can occur (unless of course they already contain integrated algorithmic execution code). Therefore the Volatility Time Bands provide an overlay to this output to allow improved execution of the orders, either algorithmically or manually.

figure 2

Figure 2 - Source: CQG, inc. © 2009. All rights reserved worldwide.

Both these proprietary tools are being used as part of the trading process by Chiron Investments. Potential entry and exit signals for alpha models are first tested in CQG's Entry Signal Evaluator. The output is then uploaded via the CQG API to MATLAB where portfolio stress testing is undertaken for both entry and exit signals. Validated models are then returned to CQG via the API where algorithmic execution is triggered using either Range Deviation Pivots or Volatility Time Bands (or both). (Depending on the markets involved, final order execution might alternatively be undertaken using third party software such as Apama or 4th Story.)

figure 3

Figure 3 - Source: CQG, inc. © 2009. All rights reserved worldwide.

A typical scenario as regards algorithmic execution sees a proportion of the necessary volume traded on the original close of the signal bar, with a fixed window of opportunity to execute in the lower timeframes, before finally shifting to the pyramid techniques. In order to smooth the profit curve, multiple timeframe technical signals that duplicate in any one hour in multiples of five minutes are used as both pyramid points or partial stop and reverse logic, such as a ten minute sell appearing against an existing sixty minute long. Volume also varies depending on what time of day the higher timeframe technical signals are triggered based on the Volatility Time Bands.

Shaun Downey

Shaun Downey

Shaun Downey

Shaun Downey is one of the founders of, which provides real time market commentaries and trade ideas in various formats, many of which can be incorporated into automated trading models. In addition, I-TRADERS distributes technical tools and scanning engines that can be used to refine algorithmic trade execution.

Shaun has been an active professional trader for more than thirty years, trading for various firms including Rudolf Wolff, Fulton Prebon and AFP across multiple markets and instruments. He is head of technical analysis at CQG as well as being a manager at Chiron Investments LLP and acting as an investment advisor for pension fund manager Gray and Associates.