None of this gets any easier, does it? We're all looking for trading opportunities, and such opportunities are getting harder not only to identify, but also to evaluate and to trade effectively in an ever more competitive environment. The challenge is not just the rarity of a really good idea, but the number of eyes that see it when - or before - you see it. If it's good enough for you, it's probably good enough for several hundred other traders to spoil it before you get to it.
So what can you do? One plan of action might be, first, to look at different ways of redefining your data. Don't just zoom in on certain times or areas of volume or dark pools; stretch your approach to incorporate changes in your visualisation of data. From that starting point, you can then begin to combine some of the more recently developed hardware technology with existing software to try and identify a fresh trading edge.
Let's look at this in more detail. When you're looking for high-frequency, multi-asset opportunities, for example, your process of redefinition needs ideally to maintain sensitivity without incorporating lag. One way of achieving such an ideal is through the use of TFlow bars. These look at the relationship between the point where the market actually traded, and the bid/ask that was available. TFlow has three variations on how actually trades or bid/ask can be viewed, and then you get a variety of graphic representations of shifts in the depth of market on either side of the current bid/ask spread.
The first of the three variations is an aggregated cumulative number of changes in the bid/ask up to a limit of 20. These thresholds build a new TFlow bar when that limit is breached. The second has an unlimited number of changes but is parameterised by a set range. The third uses a proprietary algorithm (Smoothed TFlow) that identifies trend and can redefine highs or lows for pattern recognition, even though price may not have actually traded at one of the extremes. Note that this can also be aggregated to up to 20, as with the first variation.
Figures 1, 2 and 3 show the same data in its different guises, with the histogram below identifying the actual volume hit on the bid and taken on the ask. The width of the TFlow bars indicates the relationship of absolute volume within that bar and widens or contracts and the colour indicates how much was taken on the offer (green ) and hit on the bid (red).
Figure 1 has the additional study called TFCross. TFCross displays two lines. One is the 10-bar running sum of the traded volume at the ask price, coloured green and called TFVolBuy. The second is the 10-bar running sum of the traded volume at the bid price, coloured red and called TFVolSell. Note how, in the rally, the green indicates that traders are lifting the offer, and before the price tops out the green line drops back to the same level as when the chart was moving sideways before the uptrend started.
Figure 2 shows the range based bars on the same data which is far more condensed. Testing shows that these bars are most useful when looking for range building quickly in low volume, or for reversals against the times when new highs or lows are made, and the stops routinely activated at those times fail to extend that trend.
Figure 3 is a smoothed version and tracks the trend in a similar fashion to normalised TFlow. Due to the fact the highs or lows can be simulated depending on the trend, a signal that it has ended is when both the high and low of a bar undoes the pattern of the previous trend. In this chart the trend is shown when a bar makes a lower high and low than the previous bar.
Next we look at the data type's relationship with the Depth of Market.
Figure 4 shows a visualisation of the relationship of the depth of market (DOM) for each TFlow bar. It is weighted to give the highest prominence to the nearest bid/ask beyond the current bar and calculates this with a shifted weighting up to 4 either side of that. It can then be smoothed out via a simple average through a linear or market adaptive calculation.
The DOM tracker study plots two lines: a weighted sum of the offers and the weighted sum of the bids.
The final value of the bid and ask lines is calculated at the close of the TFlow bar. It is calculated as follows:
- In Linear mode, four bid price levels and four ask price levels next to the inside market are used for the weighted sum.
- In market-adaptive mode, only three bid price levels and three ask price levels are used based on the following criteria:
- - If next bar's middle is higher - then included are bid price levels three, four, and five, and ask price levels two, three, and four.
- - If next bar's middle is lower - then included are bid price levels two, three, four and ask price levels three, four and five.
- - If next bar's middle matches the current
bar's - then included are bid price levels two, three, four and ask price levels two, three, and four.
One somewhat basic interpretation of the connection between TFlow and the DOM tracker is to use a linear regression to dictate the trend. If it is up, for example, the offer line (red) in the DOM tracker should rise sharply as new residual offers are sucked in, whilst the bid line (green) should fall or go flat as traders simply trade at the market at the offer. In effect, the apparent 'bubble' between the two lines signals the opportunity.
The trend is over when the bubble begins to deflate or the DOM tracker lines cross back the other way. Figure 4 shows that the first part of the rally is sucking in residual offers, and more bids are also being entered, thus providing an underlying fuel to the trend. However, by the time the price reaches 121.50 both lines start to drop sharply, and the end of the trend is dictated when the lines finally cross back the other way.
It is also possible to build your own custom visualisations. Figure 5 shows a five tick range aggregation of trade flow bars of the DAX with the TFlow cross custom study. The green line is the five-bar running sum of trades at the offered or ask price (buying). The red line is the five-bar running sum of trades into the bid price (selling). Notice the three consecutive lower peaks by the green line, indicating less buying pressure. The blue trend line marks a lower peak on the buying pressure relative to the second test of the high. Traders were no longer aggressively buying. Regular price bars do not show this type of information to the trader.
Additionally, with advances in hardware beginning to match the power of programmes such as Matlab, new avenues can be explored. Matlab has the power to record video and then use mathematical formulas to spot unusual behaviour. Now, obviously enough, this form of visualisation functionality has primarily been used in a completely different field; in enabling the use of cameras at prominent public places such as major railway stations in order to spot unusual activity or identify abandoned objects that could be bombs. But this is still pattern recognition, and its evolving methodology can be instructive.
The underlying concept can be transferred to the analysis of remodelled data using high performance technology to look for patterns that are beyond detection by the human eye. For example, the use of hardware such as the Nvidia Tesla GPU card (which can perform calculations many times faster than a CPU) enables numerous data images to be analysed in parallel at high speed. This performance can be further enhanced by software applications such as Accelereyes Jacket (www.accelereyes.com), which is reviewed elsewhere in this issue. This allows a different type of trading opportunity (in addition to those generated by established high frequency and systematic techniques) to be rapidly identified and exploited in real time. Same data, same markets - just sliced differently...
Shaun Downey 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 acting as an investment advisor for pension fund manager Gray and Associates. He is one of the founders of www.i-traders.com, 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.