TSCI.pngProgressive Tradescapes


The purpose of a tradescape is to get a long-term picture of a given entity's trading landscape for long term investors and trend traders. A progressive tradescape allows a further visualization of those landscapes across time.

We know that financial markets are hardly constant even across fairly wide periods of times. Historically, there are multiyear periods where financial markets are in a sharp slump, and similar periods where the markets are in regimes of sustained growth. One could readily argue that such exit for every fractal scale, from tick data to weekly data.

A progressive tradescape is simply a series of tradescapes that are sequential in time. The tradescapes for these different time segments will very clearly illustrate the sharp differences in the performance of an entity across various periods of time. The function of a progressive tradescape is not to find entities that are immune to such cycles, since that is highly unlikely, but rather to graphically convey where there is resilience or weakness, and to see how constant across time a trading system, assumed accurate in terms of trading the order in the price movements, would have been when operating at a given time horizon and with a given lag.

The AAPL Progressive Tradescape

The following is a four-plot progressive tradescape of AAPL covering nearly 16 years of its price history. This includes the massive drawdown that occurred in the aftermath of the tech bubble (the second plot) and the one that occurred following the financial crisis (across the third and fourth plots). For this progressive tradescape, we scale the RRt of the Z-axis (the contour) for a minimum of 1 so that we see only where there is more reward than pain in the trading system.

timescape01.png

The plots are encouraging, and indeed more recent 8 year time shows a greater lag tolerance than the prior 8 years. We see the resilience this entity has shown across time. If one had traded a system with an information content or time horizon corresponding with an EM length of 30 and a lag fraction of 0.9, one would have done quite well across all four periods, even with these two cataclysmic events taking place. On the other hand, if one were trading a simple signaler with a high lag, say 1.1-1.2, things would look as rosy. Let's say a trader chooses a signaling system with an EM length of 15 and a lag of 1.2 based on the first four years of this period. One would see more pain than reward in the next four years, and the weakest of what might be attainable in the next eight.

Why does a trading system signaler work for a time and then cease working? What makes a signaler robust through various market states? Where does luck factor in? An individual who just happened to choose a simple high-lag signaler with the right time horizon for the signaling might have fared very well from 1997-2000. That trader might have seen that same system fall apart in terms of performance for the next four years. The fact of the matter is that entities change. They evolve. They can become stronger, more robust, and more resilient to the vagaries of the marketplace, or they can become far more vulnerable.

The MSFT Progressive Tradescape

AAPL is an unusual security. For any extended period of trading AAPL, no matter what the system happened to be, one was likely to see some return, even if it was well below the buy and hold. For most settings, there was more reward than pain across the 16 years. For most securities, the >P (more gain than pain) scaling is too stringent for a progressive tradescape. It is usually more helpful to view a progressive tradescape in terms of >0 (any gain). The following is the progressive tradescape for trading MSFT long across the last 16 years:

timescape02.png

The concept of this choice of >0 scaling is that if there is no profitability with an EM signaler, something designed to capture all of the useful tradable order in the price movements, then one's real-world signaler isn't particularly likely to fare better without also processing the chaotic movements with some efficiency. The above MSFT progressive tradescape uses a zero RRt lower threshold. All gray areas are zones of loss instead of profit. Here we see MSFT was able to potentially return a profit for all four of the four-year periods, but the period containing the Internet bubble aftermath was especially challenging in terms of trading system signaler.

Every entity will be treated differently in the marketplace as a consequence of prevailing sentiment. It doesn't mean the entity isn't stable, well-managed, and performing consistently as a business. It merely means it is not treated in some constant manner in the marketplace. You should not necessarily deem progressive tradescapes as maps of an entity's strength or weakness in a fundamental sense so much as how it is treated in the marketplace across very different periods of market sentiment.

In terms of overall performance, we see that MSFT had a long term tradable time horizon in the first period that hasn't been repeated. It also has a vert fast tradable zone that has moved around somewhat but was present in three of the periods. This theme seems to generally hold. The lower the lag, assuming accuracy is realized, the more likely one will find this consistency or robustness across time. Higher lag zones of profitability tend to be far more variable or inconsistent across time.

In the instance of the four years covering the Internet bubble meltdown, one had to have a very good signaler to trade MSFT long for a profit. There wasn't a great deal of upside and there was a great deal of downside to manage.

The INTC Progressive Tradescape

The following progressive tradescape is the >0 (profitable regions only shown) scaling for INTC for long trading these same 16 years:

timescape03.png

Few would argue that Intel is a very strong entity technically and the company continues to push the edge. One of the items that intrigued many techies was how the industry would cope when it ran up against the barrier in physics where the line width in chips approached the width of an atom. We were accustomed to this very attractive increase in computer performance with each succeeding generation of CPUs. What would happen when the no-brainer impetus to buy new computers for that doubling or quadrupling of processing speed ceased? That did occur in the 2004-2008 period represented in the third progressive tradescape plot.

The SPY Progressive Tradescape

timescape04.png

We look at these same 16 years of the SP500 (here SPY) to see if Intel's unusual weakness in this third 4-year period was due to weakness in the overall US market. We see that is not generally the case. If anything, the overall US markets were stronger in that period.

The CSCO Progressive Tradescape

One would expect a hardware producer for the Internet technologies to have been very hard hit during the Internet bubble collapse. The following is the Cisco progressive tradescape for the past 16 years:

timescape05.png

We again look long only with a scale that shows all profitable performance regions when trading order for full accuracy. Understandably the tradescape is extremely strong during the Internet bubble rampup, and weak during its burst, but something new has emerged in the past four years. It is now quite difficult to trade CSCO on a fast time horizon, something that was not true during any of the prior 4-year periods.

The BHP Progressive Tradescape

The following long-only progressive tradescape for BHP has just 11 years of data available. Here we also look at all profitable regions:

timescape06.png

If SPY is a good picture of the overall US markets, we can find periods of strength and weakness, but across the four periods, we see no trend in terms of the entity growing harder to trade. Here with the BHP long-only progressive tradescape, we see the >0 (all profitable regions) scaling with a clear pattern where there is a huge shift in terms of optimal time horizon, and then a sharply diminished performance but a very wide and tolerant tradable zone, to where one has to either trade a long time horizon or have a very good signaler in terms of lag so as to realize a return. BHP is very much a moving target and not in the desired direction.

The AMZN Progressive Tradescape

progressive tradescapes can be useful for determining if a trading signaler is bringing its intelligence to bear when it is genuinely needed. What one looks for in designing a real-world trading signaler is to improve upon the performance of the underlying. It makes no sense to trade an entity if there is no improvement in the reward-pain from the trading system. One way to see the contribution of the signaler across different market periods is to scale each plot in the progressive tradescape so that only those areas that exceed the underlying for reward-pain are plotted. We do that here for AMZN, the >U (greater than underlying) scaling:

timescape07.png

In all four long progressive tradescape plots for AMZN the underlying RRt is lovely. For all four of the close to four-year periods, the reward exceeds the pain from a simple buy and hold for the period. A good measure of care must thus be taken when these >U types of trading signal comparisons are made. A tradescape illustrates the benefit realizable from the tradable order in an entity's price movements. Some of that reward or robust growth will come from the underlying growth in the entity and some should come from the benefit of the signaler. Those respective contributions will vary across time.

If an entity's price is consistently increasing in a wide sense across a time period, there must necessarily be less benefit from a long trading signaler since there may well be far less downside movement to filter out. In such periods, the signaler would not be expected to be much more than a support for the underlying growth.

On the other hand, if the entity's price is consistently cycling or diminishing, the trading signaler must come to the forefront. The weaker the overall performance of the underlying in a given time segment, the more one wants to see tradable order than can be realistically signaled. If such is not the case, then such entities may be tradable only during certain market states. Such entities may be candidates for the sentiment augmented signaling whose landscapes are mapped with sentimentscapes.

For the most recent four year period, we see AMZN with a superb 3.32 underlying RRt. That is more than three times the reward than pain. The reason the fourth panel of AMZN looks difficult to signal is simply that there isn't much more that a long-only signaling system can add during such a period of sustained trending or underlying growth. And we see that which can be had had requires an efficient low-lag signaler. That fits fully with expectations. When it is hard to imagine doing much better than a buy and hold, one would expect any additional benefit from signaling to be anything but easy.

What one wants to see from a progressive tradescape is a tradability of the order during the weaker periods, and ideally tradability that doesn't require heroics in terms of a low-lag high-accuracy signaling algorithm. While AMZN has exhibited strong wide-sense growth across all four periods, we see that the third panel (2005-2008) has the weakest underlying (1.17) and also the strongest added benefit from signaling on the order in the price movements. And further, the progressive tradescape plot shows that no great heroics are needed in terms of a low-lag signaler. The next underlying RRt, 1.68, is the first panel (1997-2001). During that period fast signaling did require an exceptional signaling algorithm to add benefit. The second panel (2001-2005) has an underlying RRt of 1.96, nearly twice the reward than pain. Trading the order in the price movements so as to add further benefit required either a very strong signaler or a simpler one with a very tailored time-horizon.

In general, each additional increment of reward-pain requires increasingly greater signaling intelligence. In practice, it may be easy to design a real-world signaling system that doubles the RRt reward-pain from 0.5 to 1 in backtests. It may prove next to impossible to double the RRt from 2 to 4. If the tradescape doesn't show that RRt jump from 2 to 4 at any time horizon with reasonable lag, it is unlikely that any real-world signaler will achieve it, at least any that relies primarily on trading the ordered trending in the price movements.

We come back to our original question. Is directly trading the ordered trending in AMZN possible across market states? Can we identify a specific time-horizon and lag where we consistently see additional benefit from the signaler across the different market states? This is a useful exercise. If you look closely you will see the only way to realize a strong benefit across all four periods is to have a very fast signaler with a lag of 0.8 or less. That is possible, but not easy.

Revisiting The AAPL Progressive Tradescape

Is that the norm? It certainly explains why some systems work well in certain periods and not others. Let's revisit a four-panel AAPL progressive tradescape except we will now look at the >U (greater than underlying) scaling for each panel:

timescape08.png

In fairness, we have an underlying RRt of 9.12 in the most recent four years. That stratospheric reward-pain could little be expected to see much additional benefit from any signaler. We will thus look at the first three panels. Let us assume we want to use as simple a signaler as possible, and that happens to be a fairly high accuracy signaler with a lag fraction typically around 1.0. The first and third panels have RRt values of 2.18 and 2.76, exceptional underlying reward-pain. Here we see something very different from AMZN. Let's say we design our real-world signal for a lag fraction of 1.0 and a time horizon corresponding to an EM length of 20. We see that a signal of this time horizon and lag would have added benefit in all three of the four year periods. If we had an even stronger signaler with a lag fraction of 0.9, the time horizons open up appreciably from about 15-30 EM length.

If you pause for a moment and consider that the second panel contains the Internet bubble meltdown and that period where AAPL's survival was a very real issue, and the third and four panels straddle the financial crisis period, immense differences exist in market states across the four panels. And yet the tradable order in AAPL's price movements suggest that a single signaler, not even a particularly sophisticated one, could have added reward-pain benefit to AAPL across a full sixteen years of its history. If someone would have randomly selected a signaler that just happened to have the appropriate time horizon and not too great a lag, that system would never have failed in a wide-sense across nearly two decades. This is an example where 'luck' could be a very long term phenomenon.

The SPY Progressive Tradescape

One might infer that entities representing indices or overall markets might be more stable across time. We would thus postulate that the time horizon and lag settings needed to add benefit in the different time segments would tend to be more constant. Here we look at a four panel progressive tradescape and again we use the >U (greater than underlying) scaling for each panel:

timescape09.png

Clearly we see wild differences in the underlying across the four panels, from an RRt of -0.12 in the second panel to the 3.97 in the first panel. As expected, there isn't much a signaler can do with an underlying RRt of 4 where the reward is already so much greater than the pain. Still, a very fast low-lag signal with an EM length time horizon of 10 and a lag fraction of 0.8 would have added benefit in each of the four four year periods. The signaler does well in the two panels where the RRt is weak. However there is no higher lag common time horizon where benefit accrues in all four periods.

The EWA Progressive Tradescape

Here we look at the US ETF for the Australian market, EWA":

timescape10.png

Again we use the >U (greater than underlying) scaling for each panel. We plot just the added benefit from signaling the ordered trending in the prices. If you look closely, you will see common EM lengths across the four panels from about a lag fraction of 1 and lower. Quite a few different signalers might have added benefit for each of the periods.

Revisiting the CSCO Progressive Tradescape

We will now take a look once again at CSCO with the >U scaling:

timescape11.png

This is an interesting exercise in finding a common EM length across the four very different market periods where some long signaling benefit accrued. Look closely assuming you have a 1.0 lag fraction signaling algorithm, and then a 0.9 lag fraction signaler. This gives a sense for how exceedingly hard it is for a simple signaler with a reasonably high lag fraction to work across different market states. On the other hand, robust signalers that accurately map the ordered trending with lower lags have a much greater potential for weathering different market states.