10 Market Timing Strategies, Compared

Summary

• As investors, we would all love to be able to invest in the stock market just before it goes up, and exit it just before it goes down.
• For some investors, ‘Market Timing’ is a well-needed arrow. For others, it is pure hubris.
• The goal of the present objective analysis is to evaluate the efficiency of 10 different market-timing strategies and to determine if some of them are actually worth following.

Data

Like in my previous articles, we'll try to get a statistically significant, and objective analysis. While examples like 'look how good a 200-SMA worked for Microsoft (MSFT) in 1993' or 'Amazon (AMZN) recent movements show that earnings forecasts don't work' may be amusing stories, but they would be purely anecdotal. We need to focus on a much broader simulation.

First, while the ideal would be to run tests for thousands of different stocks, we will instead focus on the market as a whole, by simulating market-timing strategies on the S&P 500. If we consider stock prices to follow a Brownian motion, then their combination is also Brownian.

Second, the simulations must span a duration as long as reasonably possible. One could argue that the stock market wasn't exactly working the same way a century ago, so we won't try to go as far. As a matter of fact we could, until recently, get S&P 500 data as early as January 3rd, 1950 with the Yahoo! Finance's API . Yahoo! unfortunately stopped the service without warnings a few weeks ago. For more recent data, we can use the wonderfully simple Tiingo.com service, but it doesn't give index prices, only securities. Therefore we'll use SPY, the largest S&P 500 ETF as a proxy. Since the S&P 500 index and SPY ETFs numbers are not exactly the same, we scale the historical Yahoo data to match. The updated data contains almost 17,000 trading day prices and is available, as a CSV file. We'll show performances both for the entire period 1950-2017, but also for the last 10 years only. The last decade is interesting because we had both a huge drop (2008), followed by an impressive but very volatile growth.

For the sake of simplicity, we will ignore both trading costs and dividends payments, which means our results will benefit from a bias toward market-timing versus buy-and-hold. Such bias seems acceptable, though, as the main goal is to make a comparison between various market timing strategies, rather than assessing the soundness of trying to time the market in general.

Time frame

The frequency at which stocks are traded has potentially a huge impact. A strategy could be successful on a very short-term basis (a few hours), but disastrous for low frequency trading (one or two trades a year). Both time horizon have their advantages and drawback.

Long term trading gives fewer data points. If we trade only once every 10 years, we may need centuries to assess the validity of the results. On the other hand, the fewer the trades, the more reliable the results: a strategy that requires to trade every hour could give wondrous results while back-testing, but turn out to be disastrous when traded for real, because of transaction costs and slippage. A TD Ameritrade study indicated than investors trade on average about 15 times a year. This indicates investors are willing to change position every few week. In accordance with this behavior, we put our re-balancing frequency at once every four weeks. This means a theoretical maximum of 13 trades per year.

Performance Measurement

How we measure the performance of a given strategy is a surprisingly controversial topic. We'll distinguish three kinds of measurements:

Returns

Absolute performance is the easiest measurement: take the final portfolio value and compare it with the initial investment. $100 invested in the S&P 500 in January 1950 become $22,994 after 67 years, a total annualized return of 5.65%.

A more time-neutral measurement is to average the returns over the whole periods. We calculate the returns for each trading cycle (which as a reminder is 4 weeks), average and annualize them. In that case, the last 67 years of the S&P 500 show an average annual return of 9.71%.

Accuracy

Accuracy ratios indicate if we were right to avoid the stock market at a given time. We can calculate the 'market beating ratio', which is the percentage of time a strategy performs better than the market, or less discriminatingly the 'not losing to market ratio', this is the percentage of time a strategy perform better or equally than the market. Two other similar measurements are the 'positive returns ratio', which shows the percentage of times returns are positive, and at last the 'loss prevention ratio', which indicates how often a strategy prevented losses.

Risks

Returns don't take the notion of risk into account. Risk is important for several reasons. First, an investor may need to exit the market at any time, and may not have the luxury of waiting 30 years before cashing out. The riskier the investment, the more likely he is to exist at a painful time. Second, investors are, for better or worth, human beings, and human beings are far from rational, cold-blooded animals. Between a portfolio shaped like a steady climb or a roller-coaster, few would choose the second option. At last, the riskier a strategy, the more likely investors will be to throw the towel and stop following it.

This means we need to focus on risk-adjusted-returns. By far the most popular is the Sharpe Ratio. I personally think it's close to worthless, because I have yet to encounter an investor who would consider making more money that average to be a 'risk', and because standard deviation doesn't take the sequence of events into account. A see-saw of ups and downs and an initial big drop with a last-minute gain can have the same variance, but would induce very, very different feelings amounts investors.

Drawdowns, which are peak-to-trough declines in a portfolio value are a better risk indication. We can measure both the maximum drawdown, which indicates the worst loss an investor could have encountered, and the averages, which gives the Ulcer index (with quadratic means). The Martin Ratio is returns divided by the Ulcer Index. Here we'll use a slightly modified version, which I call the 'Modified Martin Ratio', by dividing returns by 1 plus the Ulcer index. The Modified Martin Ratio is always computable, and always below the returns.

Another measurement we'll use is my own 'V2 Ratio'. (You may want to look at a graphical explanation here), which is particularly applicable here since we want to compare if we can do better than the market.

Strategies

Our goal here is not to be 'smart' or creative, but only to compare some popular market-timing strategies with objectivity.

50 Days EMA

We'll start with a simple and well-known idea: invest in the stock market only if the current price is above its 50-days exponential moving averages. The underlying idea is that markets always show momentum, while the moving average allows to filter out some of the noise. Obviously the length of the moving average window is crucial important. For whatever reason, 50 days is very often advised. And the results are…. not so good.

S&P 500 - 1950 to 2017	Buy and Hold	50 Days EMA
Average Return (annualized)	9.71%	5.85%
Total Return (annualized)	5.65%	3.53%
Positive Return Ratio	61.16%	35.14%
Market Beating Ratio	-	17.92%
Not Losing to Market Ratio	-	73.99%
Loss Prevention Ratio	-	48.29%
Sharpe Ratio	0.05	0.04
Maximum Drawdown	49.33%	28.56%
Modified Martin Ratio	0.40	0.40
V2 Ratio	-	-0.03
Trades per Year	-	4.28

Click to enlarge

The 50 days EMA provide some downward protection, reducing the maximum drawdown from 52% to 39%, but at a cost of reducing the average returns to only 5.85% per year. With a Modified Martin Ratio of 0.4, same than the 'buy and hold' strategy, it doesn't make much sense to follow the 50 days EMA strategy.

In the graph above, the green bars indicate when the strategy is beating buy-and-hold, that is when it's out of the market which is going down, and red when it's performing worse, meaning the market is going up while we're 'out'.

50-200 Days crossover

Day-to-day price is by itself quite a noisy signal. The idea with the popular 50-200 days crossover strategy is to compare two somewhat de-noised signals: the 50-days and 200-days exponential moving averages. Any time the 50 days average is over the 200 days indicated a bullish market in which it is worth investing.

It's simple, basic even. But it works…

Although it decreases average returns, like any market timing strategy would do, it compensates by significantly reducing risks. Maximum drawdown is only 18%. Both Sharpe and Modified Martin ratio are up.

S&P 500 - 1950 to 2017	Buy and Hold	50-200 Days crossover
Average Return (annualized)	9.71%	8.61%
Total Return (annualized)	5.65%5.25	%
Positive Return Ratio	61.16%	35.14%
Market Beating Ratio	-	17.92%
Not Losing to Market Ratio	-	73.99%
Loss Prevention Ratio	-	30.84%
Sharpe Ratio	0.05	0.06
Maximum Drawdown	49.33%	29.16%
Modified Martin Ratio	0.40	0.60
V2 Ratio	-	-0.01
Trades per Year	-	0.57

Click to enlarge

This strategy was able to successfully exit the market before it reached bottom in 2002 and 2008. It didn't do so well in 1987, where it exited just before the rebound.

5-20 days crossover

It is worth trying another strategy with shorter, popular periods, the 5-20 days moving average crossover. Results show these periods are best avoided, as they provide less downward protection while reducing the returns by a factor of almost 2. At 3.78 changes of position in a year on average, they're also more likely to generate extra trading expenses.

S&P 500 - 1950 to 2017	Buy and Hold	5-20 days crossover
Average Return (annualized)	9.71%	5.01%
Total Return (annualized)	5.65%	3.00%
Positive Return Ratio	61.16%	37.11%
Market Beating Ratio	-	14.80%
Not Losing to Market Ratio	-	75.95%
Loss Prevention Ratio	-	39.88%
Sharpe Ratio	0.05	0.04
Maximum Drawdown	49.33%	35.37%
Modified Martin Ratio	0.40	0.21
V2 Ratio	-	-0.04
Trades per Year	-	3.78

Click to enlarge

Interestingly, the accuracy numbers are still pretty good, which tends to indicate the weakness of this strategy is that it misses big market rebounds.

5-20 days reverse crossover

Since being in the market when the 20 day EMA is above the 5 day EMA performs poorly, a naive assumption is that doing the opposite should produce good performance. But that's clearly not the case: investing when the 5 day average is below the 20 days gives even worse results, with a modified Martin ratio of 0.2 versus 0.21, and a worse 'not losing to market' ratio at only 63%, as illustrated by the numerous red columns.

S&P 500 - 1950 to 2017	Buy and Hold	Reverse 5-20 days crossover
Average Return (annualized)	9.71%	4.52%
Total Return (annualized)	5.65%	2.59%
Positive Return Ratio	61.16%	24.16%
Market Beating Ratio	-	22.31%
Not Losing to Market Ratio	-	63.01%
Loss Prevention Ratio	-	60.12%
Sharpe Ratio	0.05	0.03
Maximum Drawdown	49.33%	41.59%
Modified Martin Ratio	0.40	0.20
V2 Ratio	-	-0.04
Trades per Year	-	3.79

Click to enlarge

Above 50 and 200 days average

Another popular market-timing technique amongst chartists is to consider the market bullish when today's prices are above both the 50-day and the 200-day averages. Again, performances are poor, although it does provide some downward protection, since the maximum drawdown is only 21%, less than half what an investor could have lost with buy-and-hold. Surprisingly, this strategy requires to trade quite often, with an average of 2.52 changes in position every year.

S&P 500 - 1950 to 2017	Buy and Hold	Above 50 and 200 days average
Average Return (annualized)	9.71%	5.41%
Total Return (annualized)	5.65%	3.29%
Positive Return Ratio	61.16%	37.34%
Market Beating Ratio	-	16.07%
Not Losing to Market Ratio	-	76.18%
Loss Prevention Ratio	-	43.30%
Sharpe Ratio	0.05	0.04
Maximum Drawdown	49.33%	21.20%
Modified Martin Ratio	0.40	0.33
V2 Ratio	-	-0.03
Trades per Year	-	2.52

Click to enlarge

Above 200 days average

This is provably the most basic yet efficient strategy: revisit if you should stay in the market by comparing today's price with the 200-days average. If it's above, stay, if it's below, exit. Simple, yet impressively efficient… This prevents 31% of the losses and reduces maximum drawdown to 16% while still keeping the average return at 7.8%.

S&P 500 - 1950 to 2017	Buy and Hold	200 days MA
Average Return (annualized)	9.71%	8.09%
Total Return (annualized)	5.65%	4.95%
Positive Return Ratio	61.16%	44.86%
Market Beating Ratio	-	12.83%
Not Losing to Market Ratio	-	83.70%
Loss Prevention Ratio	-	34.58%
Sharpe Ratio	0.05	0.06
Maximum Drawdown	49.33%	30.60%
Modified Martin Ratio	0.40	0.49
V2 Ratio	-	-0.01
Trades per Year	-	1.08

Click to enlarge

Above 350 days average

There's also some robustness in the moving average span, since increasing from 200 days to 350 doesn't hamper results much. The number of trades per year goes down to only 0.62 (so a bit more than once every two years), yet that strategy reacted relatively quickly, exiting shortly after the peaks of 2000 and 2008.

S&P 500 - 1950 to 2017	Buy and Hold	350 days MA
Average Return (annualized)	9.71%	7.81%
Total Return (annualized)	5.65%	4.75%
Positive Return Ratio	61.16%	46.71%
Market Beating Ratio	-	11.68%
Not Losing to Market Ratio	-	85.55%
Loss Prevention Ratio	-	31.46%
Sharpe Ratio	0.05	0.05
Maximum Drawdown	49.33%	26.82%
Modified Martin Ratio	0.40	0.49
V2 Ratio	-	-0.02
Trades per Year	-	0.62

Click to enlarge

Moving Average Convergence Divergence

The MACD, as it is commonly called, is a very popular technique amongst chartists. Here again the idea is to compare moving averages of different periods (12 and 26 days). Instead of comparing them directly, the MACD consists in computing a 9-days moving average of their difference, then compare the difference with that smoothed version. I don't want to criticize MACD too harshly (I used it myself with some success in my technical analysis days), but as a timing strategy, the results are shockingly bad. The average return is only 4.88%, while maximum drawdown is still 28%.

S&P 500 - 1950 to 2017	Buy and Hold	MACD
Average Return (annualized)	9.71%	3.88%
Total Return (annualized)	5.65%	2.28%
Positive Return Ratio	61.16%	28.90%
Market Beating Ratio	-	19.09%
Not Losing to Market Ratio	-	67.75%
Loss Prevention Ratio	-	51.40%
Sharpe Ratio	0.05	0.03
Maximum Drawdown	49.33%	37.74%
Modified Martin Ratio	0.40	0.20
V2 Ratio	-	-0.04
Trades per Year	-	4.89

Click to enlarge

For those who think the MACD is better applied to shorter trading periods, we can show that re-balancing even once a week produces similar results:

Zenvestment 30/90

About 8 years ago, I started researching my own market-timing strategy. Some basic Machine-learning optimization identified the potential in comparing two exponential moving averages, of periods 30 and 90 days. It is worth revisiting those results on a broader period.

As we can see, overall the 30-90 days strategy performs significantly better than most of the other strategies analyzed here.

S&P 500 - 1950 to 2017	Buy and Hold	Zenvestment 30-90
Average Return (annualized)	9.71%	7.81%
Total Return (annualized)	5.65%	4.75%
Positive Return Ratio	61.16%	46.71%
Market Beating Ratio	-	11.33%
Not Losing to Market Ratio	-	86.82%
Loss Prevention Ratio	-	30.53%
Sharpe Ratio	0.05	0.06
Maximum Drawdown	49.33%	26.16%
Modified Martin Ratio	0.40	0.59
V2 Ratio	-	-0.01
Trades per Year	-	0.59

Click to enlarge

PE 10

Also called the Shiller Price/Earning Ratio, this calculates the price to average earnings from the past ten years, adjusted for inflation using the Consumer Price Index. More data is available here. The idea is that if prices are too high, a crash is more likely. Once issue is that the prices fluctuated widely. The average P/E ratio since 1950 is 19, but it stayed pretty low until the sixties, when it reached a peak of 100, only to go down continuously in the late seventies, then crawling up again for 2 decades.

As a consequence, it is hard to setup a constant limit under which we'll be 'in' the market and out otherwise. We can't de-trend the numbers, because that would require to know in advance what future P/E ratios would be. Consequently, we simply calculate a moving average ratio, and invest in the market only when the current ratio is comparatively low.

Let's be clear: the results are catastrophic. Not only this strategy performed on average very poorly, returning only 2.28% on average, but it also provided poor downward protection, which a maximum drawdown still above 50%. My guess is that although a low P/E ratio makes sense from an economics perspective, it is a lagging indicator, and the market is irrational enough to not care anyhow.

S&P 500 - 1950 to 2017	Buy and Hold	Low PE10
Average Return (annualized)	9.71%	2.41%
Total Return (annualized)	5.65%	1.19%
Positive Return Ratio	61.16%	24.05%
Market Beating Ratio	-	20.23%
Not Losing to Market Ratio	-	62.89%
Loss Prevention Ratio	-	54.52%
Sharpe Ratio	0.05	0.02
Maximum Drawdown	49.33%	46.84%
Modified Martin Ratio	0.40	0.07
V2 Ratio	-	-0.06
Trades per Year	-	2.71

Click to enlarge

Conclusion

As expected, market timing is difficult, and not simple strategy works all the time (it's probable even worse for complex ones...). That being said, at least three techniques significantly reduce risks with lowering returns too much, at least from what we can deduct from the past 67 years of data:

	50-200 days EMA crossover	200 days moving average	Zenvestment 30-90 days EMA
Average Return (annualized)	8.61%	8.09%	8.44%
Total Return (annualized)	5.25%	4.95%	5.15%
Positive Return Ratio	48.21%	44.86%	47.98%
Market Beating Ratio	11.45%	12.83%	11.33%
Not Losing to Market Ratio	87.05%	83.70%	86.82%
Loss Prevention Ratio	30.84%	34.58%	30.53%
Sharpe Ratio	0.06	0.06	0.06
Maximum Drawdown	29.16%	30.60%	29.16%
Modified Martin Ratio	0.60	0.49	0.59
V2 Ratio	-0.01%	-0.02	-0.01
Trades per Year	0.57	1.08	0.59

Click to enlarge

Analyst’s Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

Comments

[Lee Bailey]

Interesting and worthwhile intro to the topic, bravo! Market timing is a very difficult thing to do accurately and consistently, which is probably why most people believe its impossible. That and successful market timers do not want more competition, so telling people not to try is a wise business strategy for top firms.

We've been building sophisticated algotrading models for more than two years at https://grizzlybulls.com. They pull data in real-time from dozens or even hundreds of indicators and analyze it using complex heuristics and machine learning to produce actionable signals or biases that can be useful in avoiding corrections and outperforming the market as a whole. As of January 2022, our top algorithmic trading model has averaged annual returns in excess of 75%.

It took us a long time to get a model that was profitable at all, even more time to build one that consistently beat the market, and it's my lifetime most proud achievement to have one returning on par with top systems from Renaissance Technologies and David E. Shaw. Before our success, we were frustrated trying a bunch of simpler systems that would only work occasionally. We realized we needed to build something much more complex, resilient and robust.

[larry jenkins]

Interesting and worthwhile article. Regarding the Zenvestment 30/90 strategy. Do you mean when the 30 day EMA crosses the 90 day EMA? If not please explain how the strategy works in more detail. Thanks.

[Scoe]

Your Sharpe Ratio numbers are not correct...or aren't scaled to standard units. How are you calculating them?

Also, I have met many professional investors who rely on Sharpe. Sortino is better because it eliminates the "upside penalty" that you refer to. But in most cases, this really doesn't matter much if you have enough data.

[Emmanuel Marot]

I wondered how long it would take for someone to notice ;-).
It's due to the way I annualized them. I didn't care to fix it since only the relative value was of interest in this analysis.

[VR51]

Emmanuel Marot -- I suggest that you try another MA approach -- 5/20 Monthly, not daily.

[Emmanuel Marot]

Do you mean a moving average of 5 and 20 months?

[VR51]

Yes, just as you used 5/20 daily MA's, put 5 & 20 MA's on a monthly chart going back 20 years.

[Emmanuel Marot]

Since there are 21 market days in a month (252 / 12), that gives a span of 105 days for the short EMA and 420 days for the long one.
Trading every 4 weeks since 1950 gives:

5-20 month EMA crossover
Average Return (annualized) : 7.90%
Total Return (annualized) : 4.72%

Positive Return Ratio : 51.10%
Market Beating Ratio : 7.40%
Not Losing to Market Ratio . : 89.94%
Loss Prevention Ratio : 19.94%

Sharpe Ratio : 0.28
Maximum Drawdown : 27.57%
Modified Martin Ratio : 0.38%
V2 Ratio : -0.02%
Trades per Year : 0.28

[Hardog]

I always like when folks talk about Mean reversion. Still trying to find that for the FANG stocks and PCLN.

[User 581032]

Individual stock prices are less likely to mean revert. They don't have to. They can go bankrupt or grow by leaps and bounds over decades. Short term mean reversion works best with sector and national market indices. Things like Mexico, Italy, the junk bond market, the energy market, or gold. All are much less efficient markets than the S&P 500.

[Longshot2]

The best mean reversion strategies I've seen do 2 things:

1) Trade along with the long-term trend. For example, if the market is above the 200-day moving average, they only go long
2) They look for short-term opportunities. Example: buy a 7-day low and sell when it reaches a 7-day high

Larry Connors and Cesar Alvarez have done a lot of work in this area. I'd avoid mean reversion strategies that go against the long-term trend, meaning you shouldn't short very often (if at all). The only other mean reversion that may work is very long term. If stocks are trading at very high valuations (like now), move to a more conservative allocations (more bonds, possibly gold) until this changes. The article I linked earlier on Shiller PE does this.

[Emmanuel Marot]

Here's an approximation of the strategy you mentioned (implementing it exactly in my code would take some work):
- Stay in in market if the price is above the 200-days average
- Get in if the price is the lowest since last week
- Sell if it's below the 200-days average and not the lowest since last week.

The results are ok. Not spectacular but ok...

Starting in 1950, trading every week:
- Average Return (annualized): 9.48% (vs 9.76% for buy and hold)
- Max Drawdown: 37.67%
- Modified Martin Ratio: 0.54
- 5.42 trades per year

Starting 10 year ago is not as good:
- Average Return (annualized): 5.67%
- Max Drawdown: 37.67%
- Modified Martin Ratio: 0.29

[Market Map]

As many boomer aged investors most likely want preservation of capital and peace of mind during significant declines, those who want to employ dynamic risk management don't need to catch "the top" or "the bottom" tick in order to produce decent risk adjusted results and compounded asset growth.
As the S&P 500 has risen 73% of the years and declined / was flat 27% of years since 1954, if one captures a portion of " the 73%" with equities and avoids portions of the most significant declines ( > -20%, 6 % of the sample ) with investment in "flight to safety" assets, then they've won half the battle.
Movements of a simple econometric measure, the LEI, has correlated with the largest significant declines over the past 60 years.
1) Move to inversely correlated assets * ,"safe" assets, or conduct portfolio hedging during LEI signaling accompanied by the price of the S&P 500 residing below it's 10 period monthly basis and
2) move back into equity based assets on a S&P 500 price cross back above it's 10 period MA.
Constructing a diversified equity asset portfolio including highest premia risk factors ( small cap value, emerging small cap, mid cap growth, and large cap value ) for the equity assets produces added value with highest and most diversified alpha with risk mitigation provided by items 1 & 2 above.
As the LEI represents a decent measure of economic activity with symbiotic relationship to earnings and industrial output, strength or weakness in the market trend goes hand in hand. This is usually coincident with easy availability or tightening of credit which has correlated with lower or higher market volatility. Economic / market trends can be likened to a supertanker; directional change taking some time and effort . Smaller market "corrections" happen, usually by exogenous, news driven events,but don't usually impact the strength of the "direction".
* the price of VUSTX above it's 10 period monthly basis when item #1 occurs instructs the investor towards allocation into intermediate / long duration bond fund; other wise cash / money market equivalents.
. . . .
tinyurl.com/lfx7zhh ( paste links into browser address bar )
tinyurl.com/n2ouhr4

[Masmon]

Market Map - great post. What you have described I believe is the best timing strategy for the average investor which is basically recession timing.

I use a similar system but incorporate some other economic indicators, principally the 12 month sma for unemployment - together with LEI they make a great system for avoiding real economic weakness in the markets.
What a system such as this does for nervous investors like my self, is it gives me the ability (and emotional strength) to ignore all the doom and gloom articles and let my investments run, knowing that there will come a time to take action..

Authors such as "Ploutos" provide the SA community with a great service by detailing where we can gain advantage in the market. Small and mid cap, embracing the low volaitlity principal seem like a great strategy..

[Longshot2]

Thanks. Where do you get historical data for the LEI, unemployment, etc?

[Masmon]

LEI
http://bit.ly/2t1Vfc3

Unemployment
http://bit.ly/2cEDxog

Awesome website from the FED St Louis, so much data available.

[OZuro]

Very interesting article; I appreciated reading it and the work you've done is outstanding. Thank you.

[RettW]

https://yhoo.it/2mf5zv5

The above link shows that similar but more comprehensive research reveals that the 13/48 day EMA is best. So you were too high (50/200) and too low (5/20) in your two point sample.
It would be nice if you could do a local search around the 13/48 day EMA optimum, including the past two years of index returns.

[Longshot2]

It's actually 13/48.5 and that sounds very curve fit--especially the .5 part. Most likely, it just happened to work well in that test. Like you implied (I think), you'd need to do a parameter sensitivity test--check similar crosses like 12/48, 14/47, etc. I'd also like to see it tested before 1991 (when that backtest started). I'd be very surprised if it performed as well when tested "out of sample."

In addition, I'd like to see it tested on indexes besides the Nasdaq and Nikkei--S&P 500, Dow, Russell 2000, etc.

[Emmanuel Marot]

Here are the results with the 13/48 days EMA:

Starting in 1950, trading every 4 weeks:
- Average Return (annualized): 5.05%
- Max Drawdown: 35.14%
- Modified Martin Ratio: 0.22%

Starting 10 year ago results are much better:

- Average Return (annualized): 7.96%
- Max Drawdown: 14.02%
- Modified Martin Ratio: 1.00

Results are similar when trading every week. I agree it smells a bit of over-fitting, I'm a bit wary of something that works great in-sample but not so much outside. Stay tuned for a future article where we'll use artificial intelligence to find the best parameters!

[Longshot2]

Also, this article shows really good results using Shiller PE or an alternative valuation method for timing:
https://seekingalpha.c...

The drawdowns look unrealistically low, though. Something isn't right IMO.

[Emmanuel Marot]

Thanks for pointing out this article. Interesting results, and worthy of a further analysis. Incidentally, it's more about 'asset rotation' than 'market timing'. Which has probably more interest anyhow. But it's also harder to make an objective assessment because, alongside the strategy, the assets you choose to rotate have a huge impact on performances.

[Longshot2]

Interesting article. Market timing strategies work better when you add some sort of filter to reduce the number of trades. For example, buy/sell based on the 10 month MA (or simply if the S&P 500 is higher than it was 10 months ago). You only check this on the last trading day of the month.

The problem with the 200-day MA is there's so much whipsawing (buy/sell trades) around it. Many people watch this MA, so there's a lot of activity when price starts to touch it. It worked much better before everyone had PCs and web-baesd charting. When Trader Vic mentioned it in his book (early 90s), it had a nice track record.

Cesar Alvarez uses a modified 200-day MA strategy. He requires the price to trade above/below the MA for 10 days before trading:
http://bit.ly/2rV8dev

[Emmanuel Marot]

Keep in mind that in my analysis, I'm trading at max once every 4 weeks, so reducing the frequency kinds of smooth out signals.

But to your point, if we take inspiration from the MACD, and trade when the signal line (200-days moving average) is above its smoothed line (10-days), results are similar but slightly better:

Avg return: 8.19%
Max Drawdown: 26.65%
M2 Ratio: 0.56

[Sanya Robert]

One of the commonly followed strategy includes volatility in the mix. Allocate certain percentages to two different moving averages and recent volatility of the underlying. It will be interesting to see the success rate of that strategy.

[bikeeagle1]

Vincent makes an interesting point, but I see it the other way. Inefficient markets might be more susceptible to one-off news driven spikes, which are the antithesis of trend following systems. It is precisely because SPY is so heavily traded and "efficient" that it is more reliable for trend following.

I found the article to be very informative. Thanks to the author. For a follow up, perhaps he could investigate various timing strategies at the portfolio level? For example, using an equal-weighted portfolio of SPY, TLT, GLD, and VNQ, what would have been the results of various timing strategies?

Thanks.

[User 581032]

I believe there is a fundamental Achilles heel in your approach to a short term trading system. The S&P 500 is the most heavily traded index by far. Daily dollar volume traded in SPY dwarfs all other ETFs. The index is heavily watched and there is heavy participation when it moves. In other words, it's a very efficient market. You will have a greater degree of success with short term trading systems if you seek out more inefficient markets.

[PVolt]

EXACTLY. CAPE systems don't work if you are in the market when CAPE is low and out when it's high. It works if you get in when CAPE is low and hold the assets and ride to the high CAPE. Usually takes almost a decade for that to play out.

[User 581032]

There are really only 2 kinds of trading systems: trend following and mean reverting. From my own experiences, I can confirm that inefficient markets are better for a mean reverting system. You take advantage of all those temporary "overbought" and "oversold" conditions. For a long term trend following system like the author seems to desire, I'm not convinced that day to day market efficiency matters, but it will smooth out the ride. I suggest the author explore mean reverting strategies using inefficient volatile markets. The results may be much better than what's presented above.