TSCI.pngReferential Tradescape Procedures

There are four referential tradescape options. The only difference between these and their non-referential counterparts is the use of a surrogate to generate the primary trading signal. Instead of directly signaling on the entity being traded, there is a signal entity, the reference or surrogate upon which the trading signal is drawn.

Referential Procedures

Because referential tradescapes represent an entirely distinct class of signaling paradigm, the Trading Sciences tradescape product segregates all referential procedures into separate functions. For each of the four major tradescape procedures, there is a referential analog:

Directly Signaled

Directly Signaled

Referentially Signaled

Referentially Signaled







Referential Tradescape


Asymmetric Tradescape


Referential Asymmetric Tradescape


Progressive Tradescape


Referential Progressive tradescape




Referential Sentimentscape


Progressive Sentimentscape


Referential Progressive Sentimentscape


We do this to prevent accidentally signaling on a surrogate when direct signaling is desired. You must explicitly invoke one of the referential procedures if you wish to explore the value of a surrogate for the primary trading signal.

Why Referentially Signal?

The best reason to referentially signal is that one wishes to trade a basket of securities using a signal derived from an overall market or sector entity. In this instance, referential tradescapes are useful for determining which entities will be favorably traded by the surrogate and at which time horizons and lag fractions that performance is realized.

Every market entity at any given point will have some measure of fickleness that makes identifying and trading the turns difficult. At some time horizons more lag will have been historically tolerated and at others that will be less evident. More generally, there may be entities where there is so little order in the turns, and so much chaos, that a very low lag signaler is needed to trade with any effectiveness. That is especially so if one is trading entities that have evidenced little historical long term growth and one hopes to generate equity growth only from trading the local trending within what is more or less constant in a wide sense.

Simply looking at correlations and betas won't answer the critical questions. Can an overall surrogate, such as an overall market index or a sector index, appreciably smoother and somewhat immune to the wild oscillations arising within any given security represented within it, generate a signal at any given time horizon and at an achievable lag that will more effectively trade the entities in question?

XLB - A Referential Tradescape Example

Let's look at some of the components of XLB, a materials sector ETF. We will do a 10 year EOD long tradescape. All of the entities analyzed will have at least 10 yrs of historical data.

The following referential tradescape is generated for XLB as the signal target or surrogate for all of the entities in the chart. In the first plot, the XLB is signaled directly-the signal is taken from XLB and applied to trading XLB. In all others, XLB generates the signal for trading the other entities.


The tradescape, when using RRt as the performance metric, shows profound differences when scaled to only show the trading landscape where reward exceeds pain. Those that show no potential for trading order for more return than pain may be in such a place for a variety of reasons. The entity may be susceptible to sharp drawdowns that are very difficult to signal. The security mat experience very sharp jumps on positive news where systems trading order have a difficult time making entries that catch the movement. There may be little overall growth in the price of the security across the ten years. The drawdowns that occur may be long sustained ones that take a long time to recover. And the most important reason may be simply that however much these entities are part of XLB, they can't really trade with its overall movements.

A great deal of information is compressed into a trading landscape. Each point of the 700+ points in each tradescape can consist of dozens or even hundreds of trades. While we cannot easily look at a tradescape plot and know why an entity like PX or MOS or EMN is so favorable, and why DD, DOW, NEM are particularly unfavorable, we can say that any signaling based on trading these entities using the ordered price movements of XLB would have had these results across the ten years of each. For those most favorable, almost any time horizon was effective and signaling lag was very tolerated. For those most unfavorable, signaling at any time horizon even at low lag would have offered very little, at least in terms of this very stringent requirement that we have more reward than pain.

RRt is defined as the long term robust CAGR (in %) divided by the average retracement from the all-time-high (in %) across this same period. If the average retracement of an entity is 20% (on average the price on any given date will be 20% off the all-time-high), we must have a 20% CAGR from the managed trading of that entity in order to have the surface drawn in the plots.

Let us say, for the purpose of basket trading, we will relax this criterion. Let us rescale from >P (show only the surface where the reward is greater than pain) to >0 (show the surface where the reward is greater than 0, where there is a net positive CAGR across the analysis period, irrespective of the pain).


The lower limit of the plotted surfaces is now an RRt of 0, so all profitable zones appear. This should not be too surprising. Survivorship alone should suggest entities that have some growth in price across a decade, if from inflation for no other reason. Of course, the tradescape is constructed to accurately map the order in these various time series, and real world signalers will not likely have this same accuracy.

There is yet another scaling of the surfaces that is useful. The >U (reward to pain greater than underlying) scaling sets each individual tradescape plot to a lower limit that equals the buy and hold RRt, the performance of the underlying instrument with no active trading. This shows where trading the order in the price movements generated a true performance benefit.


Here we alter the display to show the gradient scaling since it is different for each entity. The lower threshold for each scale is the RRt from the buy and hold across these ten years. We now start to see the fickleness of the price movements. There is nothing a trading system can do to change the truth of the underlying it must process, but it can hope to realize a certain improvement from the active management of trades. When looking at >U plots, it is important to look at the lower limit of the scale. If the lower limit is already greater than 1.0, then the underlying entity offered more reward than pain from no trading whatsoever. One would expect less from the signaling on an entity that is already in a favorable place-there is less to seek to correct and reconcile in the signaling. On the other hand, if the lower limit of the underlying is less than 0.0, then the underlying has generated a net loss across the analysis period. Any signaler can and should make a profound difference.

If we look at SHW, we see the underlying is 1.27 for these 10 years. There is a narrow time horizon zone where that is improved upon, and we see that the lag tolerance falls off rapidly. The optimum EM length is about 30 days. If we assume a fairly weak but accurate signaling system with a lag fraction of 1.0 (which XLB readily supports), we see that only certain entities derive benefit from the managed trading using XLB signals. If one were to target an EM time horizon of 30 bars and a lag fraction of 1, PX, the strongest underlying in the set, is close to where there is no benefit. NEM has a massive hole at this time horizon. EVL, another strong entity historically, is also borderline as is PPG.

While there is an art to referential signaling, certain indications can be very clear. If managed trading, at the time horizon and lag fraction expected from the signaler, cannot improve upon the underlying historically, one would wisely question putting such an entity under that kind of system. The beauty of using a generalized EM signaling technology is that many of the difficult questions can be answered in mere seconds, long before any signaling design takes place.

Why Avoid Referential Signaling?

The greatest danger to referential signaling is the possibility that the synchronization between the surrogate and any entity being traded by it could be pure coincidence. Even when looking at a large measure of historical data, the link may be suspect. Aside from this obvious need to have the surrogate represent a fuzzy reflection of the entity being traded, one that avoids so many of the sharp events sometimes associated with a given entity, the best reason to avoid referential signaling is that it may not be needed. A simpler direct signaling system may work as well.

Unless one is deliberately seeking to trade a basket of securities using one overall surrogate that seeks to represent the smooth but fuzzy movements of the different entities in the basket, it can be dangerous to use one entity to trade another. Even if one entity has strongly tracked the entity one is seeking to trade across a long historical period, and even if there is a sound reason why that should be the case, the link making this possible could fail. Indeed it may have already failed before a system is put on line.

Referential signaling does make it possible to effectively trade certain entities that might otherwise be regarded as untradable, or at least very difficult to signal. If there are harsh or wild local perturbations in the time series, strong chaos in the microstructure, a referential signaling system may eliminate the events that break a direct signaling system.

An AAPL Example


Let's say one observes that the Stockholm index is intriguing for potentially serving as a surrogate for AAPL. The country continues to be a major mobile phone manufacturer and it is an export-oriented economy. It isn't too much of a reach to assume that AAPL's business is global, and that a US index may not be as applicable.


We have a serious issue with the index, for it has only about three years of data available. As such, we look at AAPL for just this most recent period. Its tradescape represents what could be expected from trading it directly for its orderly price movements or trending across the past three years. We see that we have two very favorable time horizons, although they are somewhat narrow. We see that beyond an EM length of about 10, there is strong reward-to-pain at every time horizon.


This is the three year tradescape for the Stockholm index. We view it with the >0 scaling, and we see good agreement with the two zones observed with AAPL.


Now we look at the referential tradescape for these same three years. AAPL is traded using the Stockholm index. Using the same >P (reward>pain) scaling, we now see superb reward to pain ratios and one of the two zones is now wider and very lag tolerant. A properly designed system would be expected to fare well, if the future tracks the standing history.


There is the principle of never using a complex system when a simpler one suffices. This is a good point to examine the two tradescapes at the very forgiving 1.1 lag fraction that occurs at EM length 10. Let us assume we use a very basic moving average signaler that is very accurate, but has this poor of a lag fraction.


Let's look closely at the backtest results between trading AAPL directly (left) and trading AAPL using $OMXS30. The referentially signaled trading has 15 rather than 14 trades in the period, and a trade length of 27 days instead of 34 days. The win rate for both is in the vicinity of 65%. One is on the wrong side of the optimum (zero lag) signal 31% of the time rather than 26% for trading directly. The 3 year trend for both cases is nearly identical, a 38.75%. The r for the fitted trend is better for the directly signaled case. The referentially signaled backtest has a better RRt and Sharpe, a weaker R (which looks at worst case drawdowns). While these stratospheric reward-pain values are generally improved a bit by referential signaling, and the tradescape appears more robust, one must weigh this indirection against the risk of this link between the two entities becoming weaker walking forward or not having the same measure of constancy or efficacy across time.

Referential progressive tradescapes

While three years is too little time to do a comprehensive progressive tradescape, we can use two periods with the predefined 50% overlap to get a sense for the progression in time for each signaling system.


These are the two AAPL progressive tradescapes for direct signaling. The second period is certainly far stronger than the first. When a tradescape includes so few bars, the upper half of the surface should be disregarded as this starts to approach a buy and hold or out of market for the entire period.


The referential AAPL on OMXS30 may be consistent at the EM=10 zone chosen, but there is little robustness at different time horizons in the first of the progressive tradescape sequence.

This is a recurring theme in tradescape analysis. The higher the lag fraction, the more one has to wrestle with tradeoffs when looking at the variability across time, often to the point of having to accept extensive periods of non-performance. The more efficient the signaling with respect to lag, the more one can expect to find a greater constancy across time in terms of performance. This suggests that lag tolerance is anything but constant across time as one would expect from the influence of panic selling periods having a greater measure of chaos or lack of order. The better one's signaler, the more resilient one might expect the performance across time to be. Even with a very effective signaling system, however, one must still choose a viable time horizon that is favored by the entity being traded.

Referential Asymmetric Tradescapes

In general, referential signaling will tend to use surrogates that reflect a mix of entities averaged together as in a broad market index, a segment index, or an ETF that represents such an index. The net effect of such a surrogate should be a roughly symmetric signaling where the same information content is best used for both the entries and the exits. The different asymmetries simply tend to average out to where a symmetric signal is close to optimal.

In instances where the surrogate is not an average of many entities, but rather a single entity that has either a slow to enter - fast to exit or fast to enter - slow to exit behavior, it may be advantageous to signal that surrogate asymmetrically. For this reason, referential asymmetric tradescapes are available.


If one is trading an entity that is in general a slow to enter - fast to exit type of instrument, the optimal signal asymmetry is high. A lot of information must be processed before an entry decision occurs, but far less information is needed to make an exit decision. These are typically weaker entities that traders selloff without too much provocation when things do not go their way. These asymmetry in such cases is greater than 1, this being the ratio of entry information used to the exit information used.

Conversely, if one is trading an entity that one can safely enter quickly but where one is wise to be patient for an exit, the optimal asymmetry is low. Very little information is needed for an entry decision, but a good deal of information must be processed before determining it is actually time to exit. There are typically stronger entities where there is more of a resistance toward selloffs. For these entities, the asymmetry is less than 1, again the ratio of entry to exit information.

If one looks at a conventional range breakout signal optimization that allows for different breakout lengths for entries and exits, the resulting 3D response surface will reflect the nature of this changing asymmetry in the optimum signal. This is actually a very convenient way to get a technical picture of how markets regard a given entity. In actual fact, the optimum is likely to move across the optimization space with time. During periods of extreme negative overall market sentiment, even the strongest entity may for a time behave with a slow to enter - fast to exit asymmetry. In periods where everything is surging, even weaker entities will behave with a fast to enter - slow to exit asymmetry.

Unless one is seeking to adapt asymmetry with market sentiment across time, the best one can do is to find an asymmetry that generalizes well to all market periods. As in a turtle type signaling system with a 20 bar entry and 20 bar exit, one trades an asymmetry of 1, or a 20 bar entry and a 10 bar exit where one trades an asymmetry of 2. In actual fact, however, the optimal asymmetry is likely changing as the market sentiment varies. This is why signaler algorithm optimization in general is both difficult and suspect.

On first inspection, it would make sense that a reference or surrogate should have the same asymmetry as the entity being traded. Consider, however, that the reason such an asymmetry exists in the entity being traded may be due to the fact the time series for this traded entity contains a number of chaotic events impacting either the entry or exit events in such a way that this signal asymmetry exists. In such an instance, a surrogate that can function symmetrically may offer better or more robust performance.

Referential Sentimentscapes

It is a very advanced signaling system that operates on potentially as many as three different entities. A sentimentscape is a two stage signaling system where a slower time horizon signal is used to set sentiment states where only long or only short trades are permitted. The overall market sentiment is often drawn from a different entity more representative of the overall markets, although it can be the same entity as that being traded. When the primary signal is being generated from a surrogate, there are a host of things that could go wrong. The link between the traded entity and the surrogate could cease to work. The entity used to represent the overall market sentiment could cease to work for the entity being traded. We furnish this procedure for completeness, but recommend caution and to use the simplest and most robust signaling approach you can find. It is often the case the very issue one seeks to resolve with the sentiment filtering, that is to avoid entering marginal or dubious trades in a direction that goes against the current wide-sense sentiment, also solves some of the entry-exit issues for which one would use a surrogate for the primary signal.

Referential progressive tradescapes

In the example above, we looked at a time sequence of referential tradescapes. These progressive tradescapes are especially important when using a surrogate for the primary signal. It is a good way to check the link between the surrogate and the traded entity across time. It is possible that the very favorable result in a long-term referential tradescape comes from one or two specific time periods where the link just happened to be especially strong, and perhaps coincidental rather than causal. A referential progressive tradescape will help to identify periods where this link between the two entities wasn't in place. Equally important, a referential progressive tradescape will give a sense for how must any given link can be trusted.


Referential signaling is fraught with pitfalls. If one is trading a basket of securities using an index entity, those individual trades on each of the entities in the basket are not the same thing as trading the index directly. Some of this is simple common sense. Let's assume one is trading four entities in a basket whose breakout optimizations are [10,50], [50,10], [10,10], and [50,50]. Let us assume the optimization for a properly weighted entity combing the four mathematically is a fuzzy one whose center optimizes to about [30,30]. What might one expect from trading all four entities at this [30,30] breakout pair setting? One could easily lose on all four of the entities traded in the basket. The same logic applies to referential tradescapes. One not only wants to see the surrogate serve as an effective signal for a given entity, but if there is to be one master signal for many entities, one wants a common favorable tradescape region across all of the entities chosen. If there is a lot of scatter in the optimal time horizons and lags for trading the entities intended for a basket, referential signaling is questionable.

While progressive tradescapes can check somewhat for the robustness of the synch-up between the surrogate and the traded entity across different time periods and thus states of market sentiment, a referentially signaled system is at the mercy of this link carrying forward in time. It isn't hard to imagine various scenarios where that link would break down.

A surrogate based on an overall market or sector index is indeed far less likely to have the sharp perturbations that occur from any of the entities comprising it. On the other hand, one knows and fully expects to add a fair measure of "fuzziness" to the response. The problem is the obvious one insofar as there is no way to erase the nagging doubts that the referential link has ceased to work when the various delays that are intrinsic to that fuzziness happen.

If one is seeking to trade multiple entities in a basket using a single market index and one wishes to optimize the chance of success in that process, referential tradescapes can be invaluable. Similarly, if an entity jumps trading ranges in exceedingly sharp jumps that render one's trading systems ineffectual, and this is an entity one is required to actively trade technically, referential signaling may be a viable way to address the lack of order. In all cases, one should be aware of the pitfalls and use referential signaling sparingly and only when the simpler signaling approaches cannot be used.