Performance of GLD/SLV and GDX/SIL from 2012 to date. Our premises
To the casual observer, the Dow Theory is just a timing device for US stock indexes. However, the Dow Theory lends itself well to being applied to other asset classes. None other than Dow Theorist Hamilton stated that the Dow Theory could be applied to other markets. Hamilton, though, did not actually apply the Dow Theory to other markets. As far as I am aware, no Dow Theorist followed Hamilton’s hint.
I started this blog in
August 2012 and, based on Hamilton’s insight, I started applying in real-time
the Dow Theory to precious metals (gold and silver as a pair) and their
precious metals miners’ ETFs (GDX and SIL). Occasionally, I applied it to Chinese
stocks, and since 2018, I have applied “live” to US
bonds. I also conducted a test
with oil markets with encouraging results.
For the years 2013, 2014, and 2015, I posted the annual performance of the trades taken according to the Dow Theory. Even though the results were encouraging (outperformance of the Dow Theory versus Buy and hold – BAH- and drawdown reduction), I discontinued reporting on an annual basis, given that one year is not the adequate yardstick to compare the Dow Theory versus BAH. We should keep in mind that, on average, the Dow Theory results in less than one trade per year. Hence, in any given year, anything can happen, as we won’t have enough trades. To make a meaningful comparison to BAH, we need a sufficiently large sample size. This is why I stopped reporting on an annual basis.
Now armed with a track record of almost nine years, I feel it’s time to make an assessment.
As a reminder:
1. I am evaluating the trades as I discerned them in real-time. So what follows is not a backtest. I posted every signal on this blog and my Seeking Alpha blog as it occurred. So, for good or worse, the trades reflect my interpretation of the Dow Theory in real-time. So I am not making things up ex post facto (after the fact). Hence, it is a reliable experiment.
2. The Dow Theory “flavor” is the “classical” with Rhea’s influences. I couldn’t use what I call “Schannep’s Dow Theory”, as I didn’t have three indexes (The S&P500, the Dow Industrials, and Transportation). The most obvious candidates I found was gold (GLD) and silver (SLV) and their ETF miners (GDX and SIL). By applying the “classical” Dow Theory, I could not use some of the specifics of “Schannep’s Dow Theory”, which have been tested by Jack and I to add to performance (i.e., among others Capitulation, Bull/bear definition, etc.). Thus, I settled with the “classical” Dow Theory. It wouldn't surprise that I could successfully test in the future a combination of three ETFs (i.e. GLD/SLV/GDX or GDX/SIL/GLD, etc.) with most of the tenets of Schannep's Dow Theory. My gut tells me we'd increase performance and likely reduce drawdowns.
3. As you know, the Dow Theory is based on the principle of confirmation. To this end, we need at least two markets. In 2012, my gut feeling told me that gold and silver made a good “pair”. Both are precious metals, and both tend to act quite similarly. Same with GDX and SIL. The issue of finding the right “pair” is not a settled one for me. I still continue researching it. What makes the best “pair”? Both assets should be similar enough but not identical. If the correlation of the two assets were close to 1, then we would derive no benefit from confirmation. By the same token, we’ll derive no benefit if the two assets behave very differently (i.e., S&P500 & Copper). So finding the right “pair” is like Goldilocks, neither too hot nor too cold. The new inroads I’m making with the principle of confirmation and selecting the best “pair” is the subject for a future post. My preliminary research shows me that:
a) GLD/SLV might not be the best “pair”. Currently, I am researching betters pairs (SLV/SIL to trade only SLV, GLD/GDX to trade only GLD, etc.). We should consider lead/lag issues when selecting the pair, and we should not necessarily trade the two components of the pair. However, as we will see, results with the non optimized pair GLD/SLV have been satisfactory;
b) Any given “pair” may not work equally well with both assets. One of the two assets involved will outperform the other. Hence, as an example, GLD might work best with GDX, but GDX might work best with SIL. Lead/lag has to do with finding the proper “pair”.
c) There must be a logical explanation confirming the empirical results. If backtests showed that bitcoin worked as a good “pair” for crude oil, I’d be suspicious, as many correlations are spurious. Back to GLD/SLV, while may be not the most “optimized” pair, it makes sense and gives me confidence to trade it.
4. Given that the “pairs” used have not been optimized for one of the two assets making the “pair”, I recommend taking a 50% position in each of the assets involved. We will never be too wrong by doing this, and we will be more in tune with the underlying buy and sell signals. Exception: In the “pair” TLT&IEF, I don’t trade IEF as it has too low volatility.
5. By splitting our investing capital in half, we derive the added benefit of taking two trades instead of one (we buy 50% GLD and 50% SLV, instead of just 100% GLD or SLV). Since the Dow Theory is a long-term trading device with less than one trade per year, we know that if we stumbled across three bad trades (which may happen without the system being broken) we’d be underperforming for a long time. This is why long-term trend following tends to be unusable for most traders. It wrecks your nerves, as the scarcity of trades get you stuck in long periods of underperformance. After a couple of bad trades, it may take a long time to reach new equity highs. By taking a 50% position, each Buy signal implies two trades that add to diversification and tend to lower the time in drawdown (the more trades we take, the more likely the system's expectancy will reassert itself).
6. I will compare the performance of the Dow Theory to B&H. The proper way to comparing performances was explained in-depth in this post.
7. While the Dow Theory may be successfully applied to many assets; not all assets are created equal. The Dow Theory is particularly good at avoiding bear markets. So the Dow Theory works best with assets that have a big upside and, occasionally, a sizeable decline. Hence, the Dow Theory is an excellent trend filter when applied to high relative strength stocks. We know that high relative strength stocks tend to outperform the index but are plagued with drawdowns. We'll fare well if we remove the downside (drawdown in bear markets) while keeping the upside (stronger performance in bull markets). This assertion has been tested by me, so it is not a casual statement: Relative strength + Dow Theory does work. Accordingly, we’ll probably make more money trading SIL & GDX than SLV & GLD. The miners have more potential for significant gains than the metals themselves, provided we cut losses short.
Results for SLV/GLD
The previous post of this saga set out the premises of our study. This post and the next one are the most important that I have ever published on this blog. You won't find anywhere else the practical and profitable implications of the Dow Theory you are about to read.
Let’s first analyze our “live” trades as per the Dow Theory for the pair SLV/GLD.
The test spans from 08/22/2012 (first long taken) to 11/27/2020 (last long closed). On 4/21/2021, the Dow Theory signaled a new long position, which as of this writing, is a winner. However, we cannot include it in our study as the trade has not been closed and the final lows following the future sell have not been made. Including the open trade would bias the results in favor of the Dow Theory to the detriment of Buy and Hold (BAH).
I only assess the performance based on "Buy" (primary bull market) and "Sell" (primary bear market signals). Even though on 8/20/2020, I suggested some unloading of positions for precious metals near the top of the market and before the "Sell" signal was given (which resulted in "cashing in" profits just before the downturn set in) I have not included such trade in our comparison to BAH to our detriment. Only clear-cut 100% Buy and Sell signals have been considered.
Therefore, I feel that if there is a bias on my end, it is against the Dow Theory.
Here you have the highlights:
Total long trades: Seven round trades. The test spans. 9 years, one month and 7 days (from 6/28/2012, bear market bottom before the first Buy signal and 3/8/2021 bear bottom before the current Buy signal). Thus, each trade lasted slightly more than one year, which corresponds with the “typical” classical Dow Theory duration for the average trade, which is more than one year. On each Buy signal (primary bull market), we take a ½ position in each ETF (GLD and SLV).
Maximum drawdown buy and hold: -51.81% (for a 50% position in SLV and 50% in GLD)
Maximum drawdown Dow Theory (closed trades): -24.65%. % (for a 50% position in SLV and 50% in GLD)
Final capital buy and hold (starting at 100): 93.75. Hence buy and hold lost for an allocation of 50% to each ETF -6.25%.
Final capital Dow Theory (starting at 100): 99.81, which translates to an outperformance versus buy and hold of 6.06%, that is ca. 0.67% p.a. Please mind that the negative “gravitational” force against long positions has been powerful during the period tested, as BAH suffered a maximum decline of -51.81.
When comparing the drawdown of buy and hold to that of the Dow Theory, the latter reduced the maximum drawdown by 52.43%. This dramatic drawdown reduction confirms what I have repeatedly asserted: The primary source of the Dow Theory outperformance is loss avoidance. Hence, when markets (as US stock indices in the last 12 years) lack significant drawdowns, the Dow Theory will indeed underperform. However, “underperform” does not entail “losing”, just winning less.
Now some observations and suggestions for actual trading:
1. Please mind that my not yet finished research shows that the pair GLD/SLV was not the best combination to derive timing signals despite our outperformance. Thus, to prove my assertion, I tested the pair GLD/SLV on GLD (50% position) and SLV (50% position) with a breakup system. A breakup system is the trend following device that most resembles the Dow Theory (albeit it is like comparing a Ferrari and a Skoda). After all, the Dow Theory is based on the breakout of secondary reaction highs and lows. I tested a breakout system with the following rules.
a) Pair used to appraise trends: GLD/SLV. Assets tested GLD and SLV.
b) We buy an “n” days high provided there is confirmation (on the very same day or in the last 50 trading days). Input values from n=10 to n=130 with increments of 10 days.
c) We sell an “n” days low high provided there is confirmation (on the very same day or in the last 50 trading days). Input values from n=10 to n=130 with increments of 10 days.
d) Start capital 100 units. One half for GLD and one half for SLV.
e) To compare “apples to apples”, I selected the breakout value for each ETF that resulted in 7 round trades which is the number of trades we got using the Dow Theory. A breakout value of 80 days resulted in 7 trades for SLV. GLD needed a 90 days breakout value to produce 7 trades. As an aside, most breakup values from 10 to 130 resulted in losses for both GLD and SLV.
As you can see in the Table above, the breakout system had a loss of -13.45%, which is larger than that of BAH (-6.25%). All in all, when confronted with a difficult market and despite not using the most optimal combination of assets to derive signals, the Dow Theory was the system that performed best. Both in terms of performance and drawdown reduction.
2. Is there a better pair to derive signals for GLD and SLV? Yes, and by a long shot, at least during the period traded “live”, SIL/GDX gave better signals for SLV/GLD. In other words, true to the miners’ reputation of leading the metals, if we had just taken our trades on GLD and SLV when the pair GDX/SIL was flashing a Buy signal and exiting when GDX/SIL gave a Sell signal, we would have fared much better. I coded in TradeStation® the Dow Theory buy and sell signals derived from SIL/GDX applied to SLV and GLD (start capital for each ETF $ 50,000), and I got the following results:
So we would have made a respectable +49.57% over the period tested, whereas BAH lost -6.25%. These astounding results lead us to conclude that common wisdom has it right (at least this time): gold and silver miners tend to lead precious metals (SIL/GDX' Buy signals leading SLV/GLD's occurred 4 on four trades out of 6). I plan to deepen my research on deriving signals for the precious metals from the pair SIL/GDX. An additional conclusion jumps to my mind: It seems that when the miners go bullish (or bearish), the signal is for real and has good odds of implying higher (lower) prices for the precious metals. On the other hand, signals derived from the precious metals themselves are less credible. It seems that it is easier to lift upwards or downwards the precious metals than its mining stocks. On the other hand, it seems that if people buy (or sell) the miners with conviction, the ensuing trend is genuine.
3. We are not obliged to take all trades. What would have happened if we just had taken the Buy and Sell signals derived from the pair SLV/GLD when the secular trend appraised with the same pair was bullish? As a reminder, I assess the “secular” trend by applying the Dow Theory to weekly bars. More on the secular trend and how to use it here.
Well, if we only had taken “Buy” signals when the secular trend was bullish too, we would have had just 3 trades, and our final equity would have been 118.44 that is an outperformance versus BAH of ca. 2.71% p.a.
4. Confirmation matters (multi-market confirmation). If we had only taken trades when the pair SLV/GLD signaled a primary bull market but SIL/GDX remained in a bearish mode, we would have made a negative return of -4.46% for the period tested, which remains slightly better than BAH but worse than honoring all trades. So it seems it is not a good idea to go long SLV and GLD when its pair SLV/GLD turns bullish, but SIL/GDX remains in a bear market and is not confirming. This test highlights the powerful influence of the miners on the precious metals’ trends.
5. Finally, if we had only taken trades when both SIL/GDX and SLV/GLD were in a primary bull market (SIL/GDX confirming), our performance would have been 17.67% for the period tested and an outperformance versus BAH of 23.92% during the period evaluated.
Here you have a spreadsheet summarizing all the relevant information:
As you can see, even the plain vanilla version of the Dow Theory (take a 50/50% position on GLD and SLV according to the trend determined by SLV/GLD) managed to beat BAH. A standard breakout system fared much worse. More sophisticated applications of the Dow Theory (using secular trend filter, seeking confirmation from the miners, etc.) boosted the outperformance. Any wonder why I am an ardent believer in the Dow Theory? It works. Even when confronted with a lousy market (as it was SLV and GLD during the period examined), we can manage to be profitable when wisely used.
From this study, I draw actionable conclusions for my own trading (with a portfolio perspective). Reader do your homework.
Conclusions:
1. The Dow Theory applied to SLV and GLD outperformed BAH during the period 08/22/2012 to 11/27/2020. Drawdowns were reduced as well.
2. The same pair SLV/GLD tested as breakup system fared much worse than the Dow Theory.
3. Given that precious metals miners tend to lead the metals themselves, the Dow Theory signals derived from the pair SIL/GDX applied to SLV and GLD resulted in significant outperformance versus both BAH and the signals derived from SLV/GLD.
4. Applying a secular trend filter to weed out dubious Buy signals boosted the outperformance versus BAH to almost 3% p.a.
5. Taking trades derived from the pair SLV/GLD when SIL/GDX remains in a bearish trend, is a ticket to the almshouse (unless the secular trend for SLV/GLD is bullish). At least for the period tested, it seems the miners should be confirming SLV/GLD.
6. A trader might be well advised to using a combination of the strategies described above. It is not necessary to bet the farm on any one concept when all of them are sound.
And what about the performance of the precious metals miners SIL & GDX? Wait for the next post of this saga. Spoiler alert: SIL and GDX' performance is Dow Theory with steroids.
Results for and SIL/GDX
The first post of this saga set out the premises of our study while the second post, analyzed the outperformance of SLV/GLD versus Buy and Hold (BAH).
Today, we will see the outcome of the trades that were signaled on this blog “live” for SIL/GDX since its start in August 2012. I repeat the word “live”. The performance is so impressive that it seems hard to believe it has not been “cooked”. However, I appraised all trends in “real-time” as they were happening, and posted accordingly on my two blogs.
The test spans. 8 years, ten months and 16 days (from 5/15/2012, bear market bottom before the first Buy signal and 3/03/2021 bear bottom before the current Buy signal). On 5/7/2021, the Dow Theory signaled a new long position. However, we cannot include it in our study as the trade has not been closed and the final lows following the future sell have not been made.
As for the highlights:
Total long trades: Six round trades. Each trade lasted almost 1.5 years, which fits with the “typical” classical Dow Theory duration for the average trade (more than one year). On each Buy signal (primary bull market), we take a ½ position in each ETF (GDX and SIL).
Maximum drawdown buy and hold: -78.58% (for a 50% position in SLV and 50% in GLD).
Maximum drawdown Dow Theory (closed trades): -39.88. % (for a 50% position in SLV and 50% in GLD).
Final capital Buy and Hold (starting at 100): 77.73. Hence buy and hold lost for an allocation of 50% to each ETF -22.27%.
Final capital Dow Theory (starting at 100): 122.86, which translates to an outperformance versus buy and hold of 45.14%, that is ca. 5.08% p.a. Please mind that the negative “gravitational” force against long positions has been powerful during the period tested, as BAH suffered a maximum decline of -78.58%. Therefore, the outperformance of the Dow Theory versus BAH is remarkable. Confronted with a “negative force” of -78.58%, the Dow Theory managed to net out a positive 22.86% during the period under evaluation.
The Dow Theory reduced the maximum drawdown by 49.25% when compared to that of Buy and Hold. This significant drawdown reduction confirms what I have repeatedly asserted: The primary source of the Dow Theory outperformance is loss avoidance. Therefore, when markets (as US stock indices in the last 12 years) lack significant drawdowns, the Dow Theory will indeed underperform. However, “underperform” does not entail “losing”, just winning less.
Now some observations and ideas for actual trading:
1. Unlike the SLV/GLD pair, my preliminary research shows that SIL/GDX is a perfect pair to assess trends to trade SIL and GDX. In 2012, I selected the pair SIL/GDX because it seemed logical: both were miners and relatively similar in their price behavior but not identical. Now, armed with more background, it seems that SIL/GDX is indeed an excellent pair to derive trends. Nonetheless, I will likely unearth another good “pair” in the future. Preliminary research looks positive.
2. Was the outstanding outperformance just luck? Since SIL/GDX is a good pair to look for signals based on confirmation, I tested the same pair with a breakout system based on the following rules.
a) Pair used to appraise trends: SIL/GDX. Assets tested SIL and GDX.
b) We buy an “n” days high provided there is confirmation (on the very same day or in the last 50 trading days). Input values from n=10 to n=130 with increments of 10 days.
c) We sell an “n” days low high provided there is confirmation (on the very same day or in the last 50 trading days). Input values from n=10 to n=130 with increments of 10 days.
d) Start capital 100 units. One half for GDX and one half for SIL.
e) For the sake of comparing “apples to apples”, I selected the breakout value for each ETF that resulted in 6 round trades, the number of trades we got using the Dow Theory. A breakout value of 95 days resulted in 6 trades for SIL. GDX needed a 94 days breakout value to produce 6 trades. The breakout system resulted in a loss of -24.27%, meaning that the Dow Theory outperformance has nothing to do with “luck”. Below the specifics of dismal results for the breakout system:
As an aside, most breakup values from 10 to 130 resulted in losses for both SIL and GDX, underscoring the light-years superiority of the Dow Theory versus typical trend following devices. Hence, it puzzles me that many “trend-following” hedge funds managing millions if not billions of dollars continue to base their trades on breakout systems. I am not criticizing “breakout” traders, as indeed, breakout systems have a slight edge (google “turtle trading” or “Richard Dennis”) but are much more vulnerable to dry spells than the Dow Theory.
3. Can we improve the gains made by the pair SIL/GDX traded on SIL and GDX? Yes, there are several ways. The most logical and Dow Theory-inspired solution is to seek trend confirmation from the pair SLV/GLD (multi-market confirmation). What would happen if we only bought and sold SIL and GDX when simultaneously the trend, as determined by applying the Dow Theory to SLV/GLD, is in gear with that of SIL/GDX? When analyzing GLD and SLV, we saw that seeking trend confirmation added value. Well, confirmation works. Instead of 6 round trades for our SIL/GDX system, we would have just had 4. However, our net profit over the period tested would have been 49.96% versus 22.16% for the SIL/GDX-only system. So, performance more than doubles. We must note that trades have been cut by 1/3 (4 instead of 6), implying that our accuracy has been boosted by demanding trend confirmation. We make more trading less.
4. We can also trade SILJ and GDXJ (the “juniors” ETFs) based on the trend determined by the pair SIL/GDX. The “juniors” tend to perform better in good times and worse in bad ones. By mitigating the downside (bear markets) thanks to the trend appraised by applying the Dow Theory to SIL/GDX and demanding confirmation from the pair SLV/GLD, we would probably make more money. A 50/50% position in SILJ and GDXJ would have returned 84.24% over the period tested. Remember that over the same period, BAH lost -22.27%. Please mind that SILJ was not available at the beginning of our test, so we only had 3 trades for SILJ, whereas GDXJ had 4 trades.
Whether the trend determined by the pair SILJ/GDXJ would be even more accurate should definitely be tested. Another project for the future.
5. What if we traded SIL and GDX based on the trend determined by applying the Dow Theory to the pair SLV/GLD? We would have lost -2.17%, which remains much better than BAH. So, trends appraised from the pair SLV/GLD have some predicting power and outperform BAH. However, the pair SIL/GDX is more reliable.
6. As of this writing, the secular trend for SIL/GDX remains bearish, so unlike SLV and GLD, we cannot trade SIL and GDX on a secular trend filter. However, if we traded SIL and GDX by applying SLV/GLD secular filter, our performance would have been a respectable 45.01% with only two trades.
The spreadsheet below summarizes our findings:
7. And what about shorting? When a bear market signal is given, we close our long position and open a short position instead of going flat. I do not particularly appreciate shorting as the gain is limited (maximum 100%), and the loss is open-ended. We also have to pay shorting fees, dividends (as with GDX/SIL), and, when dealing with futures, we have to be careful with contango. I only go short when my time horizon is just a few days. However, I find it an excellent exercise to see what would have happened if we had shorted each time a bear market was signaled. After all, the only source of outperformance for any trend-following device (Dow Theory included) is the further decline that follows a Sell signal until the final bottom is made. If we had gone long and short, our final equity would have been 179.28, a gain of 79.28% versus 22.86% would have made by going only long. The maximum drawdown would have been -37.32%, slightly lesser than the -39.88% when going only long.
Can we further improve our results? Yes: If we use the secular trend filter based on SLV/GLD and take only short trades when the secular trend is bearish and long positions when the secular trend is bullish, we would have made an astounding 138.86% with a drawdown of only -6.38% (closed trade basis) with just four short and two long trades. Please mind that Buy and Hold netted out a loss of -22.77% over the same period.
Ten conclusions for this post and the previous one:
1. The Dow Theory produced outperformance versus BAH in both SLV/GLD and SIL/GDX. Drawdowns were cut by roughly one-half.
2. SIL/GDX, at least for the period tested, was an even more accurate “pair” to assess trends than SLV/GLD. I have solid reasons to believe that this is not an anomaly. The miners tend to lead the metals, and, more importantly, significant trends in the miners tend to imply that the trend is “for real” (meaning more likelihood of persistence).
3. If we combine alternative Dow Theory approaches (i.e., demanding confirmation from two pairs, demanding the secular trend in a bullish mode, etc.), performance boosts, and accuracy improves.
4. It seems that the more volatile the asset, the better results are when applying the Dow Theory. We keep the upside while cutting short the downside.
5. If we traded a combination of all the Dow Theory alternatives, we would achieve more efficient use of capital, more trades, and, more likely, reduce the depth and time in drawdowns. I have concrete ideas about Dow Theory-based portfolio construction, which I will share one day if time allows.
6. The
issue of “pair” selection deserves further study. My preliminary research
favors creating pairs that we would trade in only one asset (i.e.,
SLV would be best traded with a SLV/SIL pair, whereas SIL would be best traded
with a SIL/GDX pair). By creating several good “pairs”, we can have more trades
resulting in more diversification and more efficient use of capital (when one
“pair” is not signaling a trend, the alternative “pair” may be signaling
“Buy”).
7. I feel that Schannep’s Dow Theory, which is the very best Dow Theory “flavor” that can be found, could be adapted to other markets (i.e., precious metals and their miners) and could result in even better risked-adjusted performance. To this end, we should appraise trends based on three assets (i.e., GDX/SIL/GLD). Schannep's Dow Theory also includes Capitulation, betting on a trend reversal when prices have gone down too far too fast. The inclusion of Capitulation to our trading arsenal could likely add value but needs to be studied. Another project for the future.
8. What began as an experiment in 2012 has been vindicated. The Dow Theory is an excellent timing system (do you want to know why? Click here and here). It has nothing to do with traditional trend-following devices such as breakout systems and moving averages. More importantly, as Hamilton predicted in the nineteen twenties, it can be successfully applied to many markets; not only to US stock indexes.
9. With all the caveats about shorting, going short adds value. Those willing to increase their odds of succeeding when shorting might only take short trades when the secular trend is bearish.
10. One is not necessarily compelled to trade GLD, SLV, GDX, SIL, GDXJ, and SILJ strictly according to the primary bull and bear market signals. One can trade on a shorter time horizon (i.e., some kind of swing trading) or with tighter stops (leveraged hedge funds based on breakout systems that necessarily must have narrow stops) aligned with the primary trend as determined by the Dow Theory. I draw several actionable ideas for my own trading. And you?
Sincerely,
Manuel Blay
Co-Editor of thedowtheory.com
No comments:
Post a Comment