November 25th, 2014
Recently, I made the remark somewhere that if my DEVX V6 random trading strategy back test was done again it would achieve about the same results as the one done on November 2^{nd}. It is always easy to make such a statement. But for me, when I express something like this, I need to show some proof or at least some evidence that it would be so. Expressing it, even if I know the end results before making such a test, it might not be considered sufficient by others.
Since all the trades would again be randomly generated, each being triggered by random functions, nothing will be the same. Nonetheless, I expect the final result to be statistically close to the last test. To make my point, I used the same enhanced version of the program as illustrated in my last paper: A Donor Within.
The task is quite simple: redo the test under the same conditions and record everything.
The program is there, it's mine, so simply run it on the same 30 stocks over the past 25 years of market data. The prior test was done on November 2^{nd}, and now, November 22^{nd} it provides 3 more weeks of data in a rising market where most stock prices have been generally up. Therefore, I should expect to obtain higher performance results since the program would have continued to accumulate shares over the test duration with an ending rise in prices. The added 3 weeks of data (15 trading days) should impact on final results as also should the dispersion and randomness of trades.
The DEVX V6 trading strategy is designed to accumulate shares for the long term and to trade over its accumulative process. It has for single main task to build its stock inventory over time. One of its greatest peculiarity is that all entries are the result of random functions. No indicators used, no technical or fundamental data. The trading strategy is made to respond to mathematical equations. This in itself should open up new research avenues.
What is expected under this new test is that all the data will be different, except for the price series themselves which will add 15 days to the already respectable 6,571 trading days as reported in the prior test. This represents 0.2283% of new price data which should be considered insignificant at this scale (15 days added to 6,571). It is the same as having the number of decision points move from 197,130 to 197,580. One might have said: why bother!
Due to the randomness of the trade entries, some stocks will accumulate more shares than in the prior test while others less. And again based on randomness one should expect to see about half the stocks having an increase or a decrease in their respective inventories. In the November 2^{nd} test, the system made 1,083,818 trades, a number in this vicinity is again expected.
The why it should be so is due to the very nature of the trading strategy itself. All trades will be positioned differently, at different prices, and at different times as each price series evolves in time. Some accumulating less shares while others accumulating more, some trading less while others trading more. There is no control over that due to the random-like nature of all the entries, but nonetheless, one could still control the: do more of this, and less of that, within portfolio constraints.
In my prior test I used BA as representative for general trading behavior. Here it is again with the added 15 trading days:
F1 - BA DEVX V6 test – November 22^{nd} 2014 | |
To make the comparison easier, here is the BA chart as generated in the prior test:
F2 - BA DEVX V6 test – November 2^{nd} 2014 | |
What can be observed from these two charts is the trade placement following about the same trading behavior. Both charts are time stamped by the executing program.
The Portfolio Output
At the portfolio level the same trading behavior as shown with BA should apply to the other stocks. The trade data for the 30 stock portfolio under test was as follows:
F3 - Portfolio: DEVX V6 enhanced test – November 22^{nd} 2014 | |
What can immediately be observed is that the DEVX V6 system generated more profits. It's easily understandable. You had this huge accumulated inventory (416.94M shares) at the end of the November 2nd test that is now valued at a generally higher price. Some $1.6B was added to the portfolio. On average less than a few dollars per share being added to an average ending price of $79.49 representing a 2.09% increase while the Dow Jones index rose by about 2.41% over the same period thereby under-performing the index.
To see where the differences came from, one needs to compare the output of these two tests. The following table illustrate where the differences were between the November 22nd and the November 2nd tests:
F4 - Portfolio Differences: Nov. 22nd 2014 - Nov. 2^{nd} 2014 | |
The first, and probably the most notable point, is that the total profits increased by $1.6B over those added 15 trading days. I should have had about the same output in the two tests (within statistical boundaries). But here, simply due to the high inventory level generated in the first test, it was easy to surmise that, if prices were to go up in general, then the total output would be greater as well.
The differences are not that major. For instance, the portfolio profits increased by 4.35% which is better than the Dow Jones index for the same added 15 days. But nonetheless, a respectable increase of $1.6B in portfolio value. The total number of trades decreased by 0.53% to 1,078,084 trades. There was a 0.90% increase in closed positions (6,960 more) which had for effect to return more cash to the trading account to be recycled later into buying more shares or trading more. There was a 0.10% increase in winning trades (1,064).
The ending cash reserves increased by 7.75%. A difference of $1.18B due in most part to the increase in closed profitable positions and as a result of the different scattering of winning trades. The percentage of winning trades increased by 1.56% to 95.53%; this includes all closed and still opened profitable positions.
The November 2^{nd} DEVX V6 enhanced test had for mission to make better use of the cash reserves by generating more profitable trades and accumulating more portfolio shares, and it ended up with even more cash than before. The same phenomenon applied in the November 22^{nd} test: 73.59% of the $1.6B of added profits is cash.
The ending stock inventory decreased by 1.01% while its liquidating value increase by 1.94% ($422M). So overall, you traded a little less, accumulated a little less shares in inventory, but closed more profitable trades, most probably for more money, while having more still opened profitable positions.
Most of the stocks in the portfolio had their profits increase over the 25 year investment period (25 out of 30, or 83.33%). The CAGR increase was a small 0.23%; this might not sound like much but over the investment interval it represented a $1.6B profit increase. All the added profit was accumulated dollar by dollar over the entire 25 years.
Trades being triggered by random functions is evidenced by the distribution of stocks having an increase or decrease in the number of trades executed over the investment period. Some 46.67% of stocks (14 out of 30) in the portfolio saw their trade count increase and consequently the remaining stocks (53.33%, 16 out of 30) saw their trade count decrease. The same thing applied to the number of closed trades displaying the same numbers. Some 53.33% of the stocks (16 out of 30) saw the number of winning trades increase while the percentage of won trades increased for 17 out of the 30 stocks (56.67%). This corroborates what was advanced in the introduction: that trades were randomly generated, and these statistics are indeed quite close to randomness.
The 6,586 trading days by 30 stocks holding matrix is sprinkled with 1,078,084 trades randomly generated where you win big, and I do mean big. And yet, you are simply executing what amounts to reinvestment policies and administrative procedures.
The Back Test
When you redo a back test, even over some 25 years of market data, one should obtain exactly the same results every time. A computer program is just that, in this case a trading script, designed to do every time what it is asked to do.
However, when using randomly generated trades as in the DEVX V6 enhanced trading strategy, there is no way to obtain the same response twice, especially if some 1,078,084 trades are sprinkled and scattered all over the entire stock holding matrix; I would say not even a chance in a trillion trillion trillion years. In the DEVX V6 system, there are some 197,580 decision points where one has to decide: if yes, take the trade at the prevailing price, sell or hold on to existing still opened positions. Due to the accumulative process at work, it would be more like decisions to increase or decrease the inventory levels or hold to existing positions.
The DEVX V6 governing payoff matrix equation has already been stated as:
A(t) = A(0) + Σ(H(1 + g_{i} + T_{i})^{t}.*ΔP)
It says that position size should grow with time, and therefore, more than one trade in a single day will be required to fulfill its mandate as time progresses. Position size will start small, and will grow larger with time by recycling generated profits. All stocks will behave differently, and it is the recycling process of the generated profits controlled by the reinvestment policies that will make the stock inventory grow larger and larger with time.
Both of the tests have the same ΔP for the first 6,571 trading days, 15 rows of price data was added for this new test on the DEVX V6 trading strategy. Even if the price difference matrix ΔP is similar, all the trading points: entries and exits will be different. As if the price series were sliced and diced in a different manner, but still on the same original price data.
Another point being made in this new test is that I can't see anyone flipping shares all day generating a $1.6B profit over a 3 week trading period. But if the volume of shares is already in inventory and sufficiently high, I can easily see that inventory appreciate sufficiently in order to generate such profits. Just a slight increase on the average inventory price will do.
The objective is to reach that inventory level with time. The above equation shows how. And the new test performed using the enhanced DEVX V6 trading strategy show that it can be done.
Here are some of the time stamped charts generated by this new test using the DEVX V6 trading script:
F5 - AXP DEVX V6 – November 22^{nd} 2014 | |
F6 - DIS DEVX V6 – November 22^{nd} 2014 | |
F7 - HD DEVX V6 – November 22^{nd} 2014 | |
F8 - HON DEVX V6 – November 22^{nd} 2014 | |
F9 - IBM DEVX V6 – November 22^{nd} 2014 | |
F10 - INTC DEVX V6 – November 22^{nd} 2014 | |
F11 - JNJ DEVX V6 – November 22^{nd} 2014 | |
F12 - JPM DEVX V6 – November 22^{nd} 2014 | |
F13 - MMM DEVX V6 – November 22^{nd} 2014 | |
F14 - MSFT DEVX V6 – November 22^{nd} 2014 | |
F15 - PG DEVX V6 – November 22^{nd} 2014 | |
F16 - KO DEVX V6 – November 22^{nd} 2014 | |
F17 - MCD DEVX V6 – November 22^{nd} 2014 | |
F18 - MO DEVX V6 – November 22^{nd} 2014 | |
F19 - SLB DEVX V6 – November 22^{nd} 2014 | |
F20 - TRV DEVX V6 – November 22^{nd} 2014 | |
F21 - XOM DEVX V6 – November 22^{nd} 2014 | |
The above sample of trading charts should be sufficient to show how the stocks behaved under the DEVX V6 program. Each chart is time stamped by the software. Each stock behaved differently as each had its own trading agenda. No two stock behave the same. Yet, as a group, they behaved remarkably well, and this over the entire 25 year investment period.
This is added evidence that the trading rules underlying this methodology are worth further investigation by anyone. After all, the basic concept is not that hard to grasp: one accumulates shares for the long term while trading over the process. DEVX V6 is doing just that as it responds to the payoff matrix equation: A(t) = A(0) + Σ(H(1 + g_{i} + T_{i})^{t}.*ΔP).
The DEVX V6 trading strategy did not break down. On the contrary, it prospered going forward producing $1.6B more in profits in just 3 weeks of added market data. Knowing that all trade entries in both tests were randomly generated, the second test confirms that even if all trades had for origin the output of random functions, it would not have made that much of a difference overall and this however these trades were randomly sprinkled over the entire life of the portfolio as long as they were within the boundaries of the payoff matrix equation.
Both tests presented remarkable levels of long term performance. Even a glance at the above 18 representative stock charts should be sufficient to motivate anyone to explore this trading methodology further. The numbers and information printed in blue at the top on each of the 18 charts can be found in table F3 above.
Also note that this trading strategy is totally scalable up or down. This has been demonstrated more than once. Therefore, someone wishing to operate at a level a thousand times less than the DEVX V6 strategy would have for equation:
A(t)*0.001 = 0.001*[A(0) + Σ(H(1 + g_{i} + T_{i})^{t}.*ΔP)]
This would mean that even with a starting account of about $30k one would still have increased his/her holdings by about $1.6M during the 3 added weeks. And since the trading strategy is also scalable up, a $300M account would have seen a $16B increase over those same 3 weeks, not to mention a portfolio valued at some $370B. C'est big.
I don't know what tomorrow will bring. What could be the value of a portfolio in say some 20+ or 30+ years from now? And that is not the question. What ever trading/investment methodology used to get there might also be somewhat irrelevant. One thing I know, with a high degree of certainty, is that the future will happen, and it will happen only once.
Couldn't I just pick a singular trading strategy today and go from there, even if I don't know which strategy out of the thousands out there would do best, and this even if I have no way of knowing what the future may be?
However, I do know that this DEVX V6 trading strategy did more than quite well over the past 25 years of market data.
If a trading strategy could not survive its past, on what basis could one say it will survive its future?
In fact, what matters most is that you, as an individual, or as an organization, can get there over the long haul. And I can certainly state that there are more than one solutions. I've found through my research that there are whole families of solutions that can enable anyone to reach their long term objectives. Mine are all expressed in a single equation which can still be improved even further. It's like having many variations on the same theme. I can't say which trading strategy will come out on top, and be the best there is, but all I want is “a” strategy that could do better than most. This is what the above performed test is saying, the DEVX V6 trading strategy qualifies. I know, somehow, somewhere, someone will do better. Maybe my point would be: can you do it now?
As a side note, since all the trades are the result of random functions, anybody would be hard pressed to outguess your every move beyond randomness. Technically, even in your own trading environment, you will have exchanged the knowledge of knowing when you would take your next trade, at what price, and in what quantity. In return for relinquishing these notions, you are given table F3 above. I find it a fair exchange.
Another of my observations from these tests was that the whole thing could be improved even further, it's not a limit. It opens so many new doors for additional research. Even though all trade entries are the result of random functions, the system is only following a single equation and is governed by it. It's modus operandi is built in.
Thank you for reading this far, with my gratitude.
Guy R. Fleury
email: This email address is being protected from spambots. You need JavaScript enabled to view it.
This paper is also available in PDF format: DEVX V6 Revisited.
Created... November 25th, 2014 © Guy R. Fleury. All rights reserved.