Back in the beginning of the year, I mentioned that Mebane Faber and I were exploring updating his Spring 2007 Journal of Wealth Management paper titled “A Quantitative Approach to Tactical Asset Allocation.” (He is the Chief Investment Officer of Cambria Investment Management).

As we discussed back earlier, Meb reviewed a simple timing model — the 10-month moving average — as a signal to enter and exit various asset classes. He demonstrated a notable performance improvement across all asset classes versus traditional “Buy & Hold” investing.

Given the asymmetrical nature of market tops and bottoms, I thought that we could improve the returns of the 10 month MA sell/buy model by tweaking the re-entry signal. The description of markets as asymmetrical is based on both data and personal experience — tops take longer, seem to be more of a process, and develop over a longer time line, while bottoms tend to be more of a shorter, sharper event.

Asked in a different manner:

Assume an investor exits equity markets at the downward break of the 10 month MA; How do the following re-entry signals compare to using the 10 month MA alone? Are there any other signals worth reviewing to as a re-entry?

This is what we will be testing over the next few weeks:

 

1. TREND

  1. BUY:  Price > 5 Month MA (test 6-7-8)
  2. BUY:  Price > 40 week ma (20/25/30, etc.)

2. REVERSION

  1. BUY: Drawdown > 50%
  2. BUY: Price > 20% from 200 day SMA  (25%?)
  3. BUY: % of NYSE stocks above 200 day MA < 15% (12.5%, 20% ?)

3. VOLATILITY

  1. BUY:  VIX > 50
  2. BUY: When more than 50 of the trailing 90 trading days > 1% moves

4. SENTIMENT

  1. AAII Stock Asset Allocation 25 year mean of a 60% — BUY: Allocation > 15% below mean
  2. AAII Bull/Bear Sentiment Ratio –BUY: < 30%

5. MARKET Returns

  1. BUY: Trailing 3 year returns < 5%
  2. BUY: Monthly Return < -7%
  3. BUY: Annual Return < -10%

6. VALUATION

  1. BUY:  Mkt Cap/GDP ratio falls > 40%
  2. BUY: US Dividend Yield > 3%
  3. BUY Tobins Q: Equity Value/Book Value < 0.6
  4. BUY:  Shiller CAPE < 10

7. ECONOMIC (Data to 1948)

  1. BUY: GDP falls > 2% year-over-year
  2. BUY: Coincident-to-lagging indicator falls 10 points from highs

8. INSIDERS

  1. BUY: Insider buying > 20% of 5 year mean
  2. BUY: C-level insiders (CEO/CFO etc) > 15% 5 year mean

9. EARNINGS

  1. BUY: Earnings < -25% Year-over-year
  2. BUY: Earnings revisions > -9% (negative 9% or worse)

 

Based on some preliminary research, suggestions from friends and colleagues, and a dollop of gut feel, this is what we are going to be back-testing.

I would appreciate any suggestions, comments or ideas on the subject.

Category: Investing, Quantitative

Please use the comments to demonstrate your own ignorance, unfamiliarity with empirical data and lack of respect for scientific knowledge. Be sure to create straw men and argue against things I have neither said nor implied. If you could repeat previously discredited memes or steer the conversation into irrelevant, off topic discussions, it would be appreciated. Lastly, kindly forgo all civility in your discourse . . . you are, after all, anonymous.

54 Responses to “Re-Entry Signals Following 10 Month Moving Average Exit”

  1. A7L-B says:

    Is there some compelling reason to use a 10 month MA instead of the way-more common 200 day MA?

    Also, some months back you solicited from us some favored technical indicators.
    Will you be sharing that info with us?

    Thank you…

    ~~~

    BR: 200 day is too whippy — lots of false signals

  2. A7L-B says:

    10. Volume

    Buy when you see accumulation days. Stay out if you don’t.

    It’s interesting that during the greatest bull market in two centuries, TA fell out of favor; Use of TA will likely become widespread before this depression concludes…

    ~~~

    BR: How do you define “accumulation days?”
    Can you express that in a formula that can be tested?

  3. Singmaster says:

    I used to use monthly MACD, basically 12 month moving average.
    But that stopped working. Now using weekly MACD.
    That would have you buying SPY Sept 2010, selling Mar 2011, buying Oct 2011, selling last week.
    I wait for 3 weeks confirmation of trend on buying and selling after one week of downturn.
    Also put toe in when Bulls drop to 15 and/or Bears rise to 55 on tickersense.
    But that’s just me.

  4. bmoseley says:

    back testing often works better than dealing the future or present.

    ~~~

    BR: I think you are referencing Postdictive Error — thats why we are laying out a premise to be tested, rather than tweaking and form fitting a model to optimize it for past performance.

  5. drctypea says:

    are you testing these signals one by one or in certain combinations? what time period(s) are you testing over?

    ~~~

    BR: Singly and depending on the results, in some combinations. As far back as the data for each goes (but no further than 1920)

  6. A7L-B says:

    So, the 10 month MA appears to be at 12335 basis Dow 30, and at 1292 basis SPX. WTF? You’d still be long!

    Using the Chimp Indicator described here a few months back, you would have been out as of 05AP12, and combined with the obvious volume data, being 8 distribution days and two accumulation days since,
    you’d still be out or short, and you would have saved youself and your clients ~6%.

    Continuing to use the 10 MMA / 200 DMA will let you lose another ~5% before getting out/going short.

    The technical case for revisiting the 3-2009 lows is compelling; much prudence is warranted.

  7. craig.r.jackson says:

    Just a quick glance of the ten month avg looks to me like it buys too late into the 2002-2006 bull market. It gets you out near the top of the 2007-2009 bear market. It gets you in a little late to the 2009-? bull market. Then does a trade flip at the recent highs, selling then buying…still long. I’d be leery of moving averages because they lag. Don’t waste your time.

    ~~~

    BR: Exactly — the goal is a better entry point

  8. [...] buy-n-hold strategy with a market timing entry and exit. Read exactly what they will be testing here. [...]

  9. Scott Teresi says:

    One idea I’ve always been interested in has been the timing signals in William O’Neill’s Investor’s Business Daily “Big Picture” column. I think these are largely mechanical. How would the 10-month SMA perform if you re-entered the market after a reversal or big up day on high volume, plus a second big up day within 4-10 days after the first? (Not sure if I have that right.)

  10. JesseLivermore says:

    I don’t know if you keep track of the names of various posters and what they have to say. If you do, you probably know me as a jerkass who butts in to nitpick you and your writers from time to time (especially Peter Boockvar’s obsession with inflation…). Anyway, I just wanted to change it up and say that this is the most valuable post I’ve read all month, anywhere. I had no idea that there was quantitative historical evidence for a simple trend-following strategy.

    Next steps? Well, now that you’ve established that this approach can be successful, you are statistically “allowed” to try many different moving-average periods to optimize your results. The original paper suggested that a 12-month MA gave the best results.

    One other thing you might investigate is the market’s tendency to “stick” near round numbers. The Dow was stuck near 1000 from about 1964-1984. Through that entire period it never deviated more than 40% from 1000, and mostly stuck in an 800-1000 band. Between 1998 and now it hasn’t deviated more than 30% from 10,000 (with the exception of April, 2009, which was a major buying opportunity). A reversion-to-the-mean strategy would have been very effective during these times. It would be quite interesting to see if other major indices (and individual stocks, as well) have a similar tendency.

    ~~~

    BR: Jesse — I write the blather I do because most of the time my model has me either in or out of the markets, with little else to do. They major reverses come along infrequently (once or twice a decade) while the minor ones once a year or so.

  11. boveri says:

    I fear there will be either tears or disappointment at the conclusion of this test if it is solely used to determine buy and sell points. There are additional prereqs for consistent success.

    First, a ferociously abiding interest in the markets. Then an intense commitment to daily study of all factors that drive market behavior. Finally a focus on reading the daily price action, never straying far from the paths it leads you.

    ~~~

    BR: Thanks for the cliches.

    I’m gonna bet you havent read much of what I’ve written on the subject and that you DONT run money . . .

  12. milktrader says:

    Meb Faber gave a keynote on this topic at R/Finance in Chicago last year. I was pleased to have seen the presentation. The strategy has been implemented in R here: http://www.rinfinance.com/agenda/2011/BrianPeterson.pdf. Also, I may add that your metric of volatility is based on a lagging indicator. There is much research (dating back to Engle) to support the notion of predicting volatility with GARCH models.

    Not meaning any disrespect, but your panoply of filters is naive. If you want to fit every point in history, you can I suppose. But (with due respect) you will find that there is no trade that meets your criteria.

    ~~~

    BR: Perhaps I failed to clearly explain what we are attempting: The concept is not to use EVERY metric; rather, its a simple test to see if any single (or combination) of these data points generates returns that are superior to the symmetrical 10 Month MA Entry/Exit approach of Meb’s original paper. We are assuming a 10 Month MA Exit — do any of these consistently outperform the 10 Month MA Entry?

  13. Liquidity Trader says:

    Dude, this looks awesome. Please post the results when you complete it.

    From your impatience at some of the imbeciles above I sense you already have a good idea what the results will look like, and they probably rock.

    ~~~

    BR: We are planning on publishing a white paper when we are through. And yes, I have a sneaking suspicion as to the results . . .

  14. milktrader says:

    Ah, in that case I vote for the simplest approach, which entails implementing Faber. Any filters should be predictive in nature. Whether that’s GARCH or some sort of kNN or svm implementation. These raw data points are interesting only so far as they are parameters in a model (or whatever you want to call it http://www-stat.stanford.edu/~tibs/stat315a/glossary.pdf)

  15. hawaiianwaverider says:

    BR, you are pushing Occam’s razor for sure but welcome to quantwork. How many of the 9 categories are needed for a buy? If you optimize all these variables you will have curve fit and while they are each good variables, how are you going to assess their contribution?

    Mebane’s work is good, simple and provides a good starting point so we are looking forward to see what you come up with. Show backtests please.

  16. louis says:

    I dont run money and all this Math makes me dizzy. This is the equation I get LTCM=BEAR.

    I’ll be in the back worrying about all the dead homeowers if you need me.

  17. A7L-B says:

    “BR: 200 day is too whippy — lots of false signals” — No, not when context is considered. How do you define the 10 month MA? This metric is rarely used or discussed. The ‘imbeciles’ would like to know.

    ~~~

    BR: The 200 day changes daily. Look at any index and count the number of signals above and below. If you read the original paper in the Journal of Finance, you will see that the 10 Month changes once a month (duh) and that intervening 30 days rarely gives you a buy/sell/buy signal over a short period of time.

  18. A7L-B says:

    “BR: How do you define “accumulation days?””

    I don’t; the industry has agreed on the definition for many years. Your question is fairly astounding; of all the indicators in the universe, volume is second only to price in terms of importance.

    ~~~

    BR: Let me refine my question: How do you express accumulation days as a mathematical formula that is a buy signal? It needs to be an algorithm that can be tested. (I assumed in the context of this discussion that was understood)

  19. boveri says:

    Rebuttal — You are correct I don’t run money, but that’s about it. These tests you are running will be more than a waste of time if they take precedence over the daily price action.

  20. CultOfMoney says:

    I think that you’d have good results with using some of the breadth indicators. Something like the McClellan oscillators (summation index) that fall under a certain trigger and then show some type of upturn, either sequential weekly or back above some trigger point. Something like the % of the SP500 under the 200 dma may also work with a similar trigger setup. Looking forward to seeing the results!

  21. RickyRoma0 says:

    I would think you could also consider: 1. Extreme market breadth signals. 2. Seasonality. 3. Fed actions

    ~~~

    BR: Breadth is a great trading signal, it would be interesting to see it tested as a longer term investing signal. Fed rate Yield curve was part of our original 20+ signals, but we wanted to get this down to 10, so it was eliminated. Seasonality looks too short term.

  22. Pantmaker says:

    Let’s keep this simple. Since 1995 there have been three buy signals and two sell signals (soon to be a third) using some definition of a sustained cross of a 10 month MA as the trigger. Six trades in 17 years. Could the entries and and exits have been better….Meh…maybe a little. The most obvious improvement to increased returns is to not just exit the long position when the trigger executes but to actually take a short position. Not shorting is leaving money on the table. If you stick to shorting indices you can ignore all of the anti short Easter Bunny bulllsh.

    Secondly. Monthly MAs are already obsolete. The big peaks and valleys of the past 17 years are over. We’ll get one last big ass-tanking and then a long boring grind down until the last of the boomer retirees sells his final share of stock.

    ~~~

    BR: Sounds promising, but everything we look at I want to see backtested through the 1966-82 bear as well

  23. Scott Teresi says:

    How would FusionIQ perform, re-entering on a Buy or Neutral signal for SPY? (assuming it had gone to Sell by the time the 10-month SMA did)

    Maybe a proprietary system like yours wouldn’t fit with what you want to publish, but I would love to see if there’s some way the two systems might work together!

  24. adrian.who says:

    Hi Barry

    I know I may be wrong assuming you don’t already know this, but still:
    http://advisorperspectives.com/dshort/guest/John-Carlucci-Best-Indicator-Ever-Update.php
    (The author admits it could not be used mechanical between 2003-07.)
    Btw, it signaled ‘sell your longs’ yesterday, May 16.

    And talking about testing different strategies over long periods, what data set would you recommend if I am interested in company level strategies, not market index strategies? From what I read, there is S&P’s ‘Point in time’ database, but I need a free one

    Can’t wait to read about your research results

  25. mathman says:

    Here come the food shortages:

    http://af.reuters.com/article/commoditiesNews/idAFL4E8GH04Y20120517

    “U.S. wheat futures climbed on
    Thursday, extending gains from the previous session when the
    grain posted its biggest daily price rise in 6-1/2 weeks, on
    concerns about dry conditions damaging the winter wheat crop.
    Chicago Board Of Trade July wheat futures rose 0.74
    percent to their highest level since May 1 after soaring 5
    percent on Wednesday, as weather concerns rattled the market.
    CBOT July soybean futures rose 0.56 percent, having
    climbed 0.6 percent in the previous session, while CBOT corn
    futures opened flat after rising 3.8 percent on Wednesday.

    FUNDAMENTALS
    * Investors covered short positions as the winter
    wheat-growing U.S. southern Plains looked to stay mostly dry
    this week, traders said.
    * Traders also worried by Russia drought, which analyst
    warned on Tuesday could cause irreversible damage to some crops
    in the south.
    * The U.S. Department of Agriculture said exporters sold
    900,000 tonnes of U.S. corn to China, including 180,000 tonnes
    of old-crop supplies, even as China said it expected a
    record-large domestic corn crop in 2012.

    * The China National Grain and Oils Information Center
    projected on Wednesday a 7 percent drop in Chinese soy
    production this year, a reduction that would boost imports by
    the world’s top buyer of the oilseed.
    * China’s corn supply is expected to stay tight in 2012/2013
    (Oct/Sept), leading to a continued increase in imports of the
    grain used mainly for animal feed production, according to
    estimates by an official think-tank. The China National Grain
    and Oils Information Center said China is expected to consume
    199 million tonnes of the grain in 2012/2013, while domestic
    production has been projected at 197.5 million tonnes for the
    year. ”
    (there’s more)

  26. adrian

    I have written about % of stocks over 200 day moving average on NYE, SPX and other indices.

    The problem is it failed in 2009 as a signla in 2008-09

    I know they way Carlucci uses it is in combination with three other metrics — we are testing a simple input before trying multiple (4) combination signals

  27. RothcoUDipthtick says:

    It’s funny, I’ve also been working on a quantitative way of expressing tactical asset allocation, but I haven’t been using technical analysis indicators rather a combination of fundamental and risk ones and spreads thereof. I still have a bit more work to do, but happy to share once done. Initial results are superior.

    Re: tech indicators – this may be a bit short term for you – but I think that a 13,34 EMA works pretty well on equity indices – picks up the intermediate signal pretty well. Considering a lot of risk on risk off phases last 6-8 weeks….

    Re: breadth you could try the Mclellan Oscillator too

    Cheers

  28. RothcoUDipthtick says:

    As a p.s I would add – how will you be determining the proportion by which you will buy/sell?

    In the method I have used, I have tried not to trade 50% of the time to reduce dealing costs if the observations are broadly within 1 std deviation…. this leaves you trading 75%+, 25%- or the outlying quartiles. Also applied a +-15%benchmark equity allocation range around the portfolio to limit tracking error.

    I guess you have more freedom, but still would be interested as these assumptions/constraints also determine performance….

  29. dhernquist says:

    I’m as big a Meb Faber fan as anyone, he shaped much of my thinking as I evolved from trading to allocation. This approach is great because it answers what you want…”What is the essence of trend here?”. BUT you are also placing complete emphasis on one day’s close, whether 10 months is the right lookback, and simply whether it closes above or below that level on some random month-end Friday. We haven’t tackled the magnitude of the above/below or where it closes on the day before or day after.

    With so many new adherents to his approach, you can choose to incrementally be in or out by breakinginto 8-12 pieces and giving each a % weight in the decision making process. Could be 1/3 for Thursday, 1/3 Friday, 1/3 Monday. Inside of that, could mandate that it needs to be X% above or below to qualify as an exit or re-entry. Inside of that, could use the 8 and 12 month to confirm that it’s the essence being quantified and not just a highly followed make or break level.

    If a break in the 10 month was like a break of a high/low requiring a stop, I’d say immediacy is better. But since moving above or below tends to be part of a methodical directional shift, there is time to act methodically in my opinion. I also think that, given month end/beginning tendencies for bullish action, postponing the decision by a week can be a discriminating factor as well.

    Love the topic…re-entry is the fuzziest part and most difficult part of this approach

  30. RothcoUDipthtick says:

    Oops – think it skipped my previous post. Just to recap, I have been working on something similar except using fundamental metrics and/or various measures of risk and spreads between as a trading system. Broadly the idea being volatility responsive allocation… I’m nearly done but would be happy to share when complete. Early results are very promising.

    As a tech indicators I have found that 13,34 EMA works well as an intermediate signal. Considering a lot of risk on risk off phases last 6-8 weeks.

    Re: breadth indicators, you could try Mclellan Oscillator that also works well….

    Cheers

  31. VennData says:

    As your control, pick a 70/30 equity/bond asset allocation.

    The 70% equity should be split 50/50 US and international index. The bond index 50/50 TIPs and US intermediate bond index. Re-balance these four asset classes annually, quarterly, monthly. It’s not “Buy and Hold,” …it’s NEVER “Buy and Hold.” It’s always “Buy, Hold and Re-balance a fixed asset allocation.”

    The costs here are much lower. The market tells you when to buy and sell. You want one more idea go 60% equity 30% bonds and 10% REITs and split the equity a quarter each US total stock market, total international market, small cap value, and international small cap.

  32. Jim67545 says:

    One thing I use is a 5-10-20 day MA to ascertain the short term direction of the market. Right now both QQQ and SPY are out of “long”, past “neutral” and into “short.” Of course, this is way more fidgety than something based on a 10 month or whatever longer MA. This method, based on backtesting, has been useful on the long side and not as trustworthy on the short side. When the method switches from one result to another I begin looking at other things and usually wait a few days for confirmation. Between trailing stops being hit and this I’m short now (ignoring higher dividend equities.)

    For individual stocks I find the Mass Index, with the indices tweaked, the most dependable single measure of several I use. It seems OK for 3rd world or smaller portfolio ETFs but not as good with big ones like SPY, I think because there is more diversity within them. It signaled a weak change in direction (upward) of SPY on 9/29/11 and again on 11/16/11. It signaled (downward) on 2/15/12 @ 134.55. SPY rose until 4/2 @ 141.84 and is now 132.83. Some stocks have busy or contradictory MIs and I stay away from them – but I’m not especially proud of my record in equities so caveat emptor.

  33. bobnoxy says:

    Back testing is great, just as long as the past repeats perfectly and all the variables remain identical…or you could just update your Ouija board.

  34. KV says:

    Barry,

    A quick comment on Faber’s method:

    Any such system will provide delusionary gains on paper studies. In real life, one does not have the luxury to stay invested in a single class (even S&P 500) for 30 years or more unless one is endowed with multimillion dollars to invest.

    From mathematical perspective, such systems must be tested for different starting points, selected at random, to provide a spectrum of likely outcomes. I have modeled many such systems which look exceedingly good on paper but when implemented one would find falling somewhere in the spectrum and when one losses 40% or more, logic and reason go out of the window.

    Could moving average be a good indicator? Mostly yes. However, 10 month average is too slow for most modest portfolios. If you have multimillions of dollars, remember Bill Gates’ line: a billion here and a billion there!

  35. CANDollar says:

    Last year S&P did some research over 40 years (Dec 31 69 to July 8 2011)

    The 65×200 EMA had the best return. Its CAGR was 7.2% vs 6.6 for the SP500. Annualized it beat the index 48% of the time. Its best year was 32.2% vs 34.1% for the index. Its worst drawdown on an annual basis was 10.7% vs 38.5% for index.

    Why are there no crosses in this research?

  36. ExpatAmerican says:

    Dumb question from a newbie investor:

    1. Did I get the gist of this right with this example – One of your buy signals triggers when Price > 5 month MA?
    BUY: Price > 5 Month MA (test 6-7-8)

    2. One of your sell signals triggers when the Price goes below 5 Month MA?

    Thanks!

  37. peflp says:

    This sounds a lot like The Ivy Portfolio approach. They use either a 10 month or 12 month MA. It works. But can it be improved upon? Yes. But I think the indicators listed above are too numerous and redundant. Also, you are missing what in my experience has been one of the best indicators for catching bottoms, the New High / New Low Index.

    BTW, anybody feel things are a little too pessimistic in the market right now?

  38. cognos says:

    I’ll throw my 2 cents to distract from this market carnage a bit.

    So the main problem with “back testing”… is getting something that worked nicely… but now doesn’t work. More history tend to exacerbate this problem… as markets 20yrs ago, tend to be irrelevant today. Things change. AND your data gets bad (everyones does).

    And then specifically avoiding something with a simple, but enormous flaw (5-yr normalized PE… misses the 96-00 run, out-of-business).

    The people I know doing this stuff well… a) they use fewer technical indictors and simpler ideas; b) they combine them with economic indicators – curve, oil prices, leading econ indicators, corp spread relationships. c) they might also combine them with seasonals. For me… rate curve alone is a pretty good indicator and place to start. Inversion = coming recession. Good slope = recovery.

    Its easy to look for the “range trade” recently… bc it would’ve worked nicely the last 5, 10, even 15-yrs. Its important to think about how the range trade fails… bc its fails huge, mainly by missing the upside. (Or buying Greek Equities… but I don’t think the US has that problem). Your search for “entry / exit” point has an inherent “range” trade bias.

    Finally… the better people seem to use the idea of “lags” more effectively. That is, interest rates stimulate with a certain # months lag. Oil prices stimulate with a # months lag. Etc.

  39. bobnoxy says:

    I think he’s reading the wrong Faber. Marc Faber’s book Tomorrow’s Gold lays it out better than anyone else for me. Find the secular trends and ride them out. That means bearish on stocks and bullish on gold. Simple buy or sell signals can help ride through corrections like with gold now. But stocks are a big loser for well over a decade.

    Look at the Nikkei for what’s possible. I know, it’ll never happen here, but that’s what they said there too!

    Look at more than just the domestic markets. Check how far down the major international markets are from their 2007 peaks, and then from their all time highs. As the right Faber showed, avoiding stocks has been, and will be the right play for years to come. You can trade, but most can’t for very long.

    Forget technical signals. How many hedge fund black boxes are working on formulas for that now like the one mentioned in the piece? Would BR’s formula be that much different from theirs, especially after back testing? Can you beat them at that game with all their resources?

    Why bother trying? Find the secular bull market and ride until it dies. With all the debt and money printing, gold has a long way to go.

  40. cognos says:

    Bobnoxy…

    But Faber would’ve said the same thing in 1981… Gold went down 50%… stocks up 500%.

    This is why he is still on TV. Not that successful.

  41. bobnoxy says:

    Do you know that he said that, or are you just posing a baseless, hindsight-enabled scenario to agree with your premise?

    I’m guessing he said that in 2001, and since then, gold is up 600% to the S&P 500′s loss of 14%. But hey, if you want to live, and invest like it’s the 80′s again, no sweat here. If you want to see who can sail the fastest into the wind and stick with stocks, knock yourself out…literally.

  42. bobnoxy says:

    Oh, and don’t forget, that 14% loss is helped by a big survivorship bias that assumes no one invested anything in Enron, Bear, Lehman, AIG, Countrywide, Citi, Merrill, Wash-Mutt, MF Global and all the dot coms that busted out too. What’s the true investor performance with those included?

  43. Scott Teresi refers to O’Neil’s follow through day concept. My understanding of the FTD rules are:

    1. Day 1 of a tentative rally begins when a major index closes higher than the day’s low or the previous day’s close.
    2. On the fourth through seventh days of the rally there must be a close at least 1.7% higher than the previous close.
    3. On that day, the volume must exceed the previous day’s volume.

    I did my own backtesting of these rules and did not find them sufficiently consistent to be reliable. http://www.breakoutwatch.com/newsletter/archive/2011-04-02.php#6

    Further analysis showed that a combination of stochRSI (price based indicator) and Acc/Dis (volume based indicator) was promising and is now used as my downward trend reversal indicator. As Barry observes, tops are different to bottoms and need different methods for identification.

    Thanks for the blog, Barry, I never miss a day of it.

  44. fdhin says:

    I did not backtest this very far, because I dont have the means. The idea is to spot the cyclical upturn:
    - When spot price of copper pierces upwards through it’s 26 week MA, then a bottom in SP500 is in or due in short order.

  45. MZeigler says:

    This is one I’ve used that combines variance with mean reversion. It should be relatively simple to test (I did it in excel with the Shiller data.).

    Using monthly returns for the index, calculate the standard deviation over rolling 10 month periods. Next, calculate 2 standard deviations below the 10 month sma. Buy when the market is 3% or more below the 2 standard deviation value for 2 consecutive periods. I know that’s a mouthful, I’d be happy to clarify if need be.

    This is a great effort, thanks for taking it on in an open format like this.

  46. Poter says:

    Monitor the mood as it shifts to “risk-on”. This is accompanied by rapid appreciation of open economy exchange rates, e.g. South African rand, from low levels. (Not tried.)

  47. Ethan says:

    Back when I had time to think about these things I ran up against asymmetry as well.  I looked at two things which haven’t yet been mentioned.  First, simple moving averages for sales and exponential moving averages for buys.  Also I looked at using crossovers.  Exponential crossing simple for buys.  Simple crossing simple for sales.  As the exponential is weighted towards the more recent price data, it reacts faster, potentially solving for asymmetry.  I also looked at secondary signals to generate if/then buys and sales.  In particular,  momentum or volatility oscillators (MACD, %B, ATR,etc…).  I would look for divergence or shorter term (but not too short compared to the subject term) momentum changes to prequalify a buy using the moving average strategy outlined above or to scale in a buy and sell.  Unfortunately divergence is hard to quantify using coding language, but using moving average crossovers on the oscillators themselves may work for pre-signal buy and sell idea generation.  Unfortunately I ran out of time and endurance to test these theories, maybe in the future.  Looking forward to the white paper!

  48. Ethan says:

    After reading my dimwitted grammar on my previous comment, I want to clarify that I looked at using shorter term secondary indicators relative to the longer term primary indicator because they don’t show up frequently on longer term price charts (especially divergence).  This is easily observable on monthly and weekly charts.  So if I were using a monthly chart for a primary indicator, my secondary indicator may come from a weekly or even daily chart.  Sorry if this was not clear.  Commenting from a iPhone in a moving vehicle presents challenges.  :)

  49. MBJ says:

    Could an element of Norman Fosback’s seasonality strategy be added? Specifically this rule:

    “Buy at the close of the third-to-last trading of each month, and sell at the close of the fifth trading session of the following month.”

    So if you do not have a position and the fourth to last trading day of the month closes above the 10 MA enter the trade on the 3rd to last day.

    If you have a position and the 10 MA is breached on the 4th trading day sell on the 5th.

    http://marketsci.wordpress.com/2010/03/09/fosback-seasonality-strategy/

  50. TomC says:

    One thing I might suggest is testing some of these on a compound basis – take three indicators I like:

    Long – if Close > 200 MA OR Golden Cross (50 over 200) OR 252 Day Momentum positive
    Short – If ALL of the conditions aren’t met.

    That way you are biased on the long side as long as any of your good entry indicators are positive. I’ve found this kind of approach may have some potential in addressing the asymetrical nature of Long versus short.

    On a side note to those who like the 200 day MA as opposed to 10 month: one way to smooth this a bit is to only enter long after a down pulse (e.g. RSI(2) 50) – waiting for the reversion that occurs after a run to the 200 day.

  51. Scott Teresi says:

    I second MBJ’s suggestion to shape entry and exit points around turn-of-the-month seasonality.

    Norman Fosback’s seasonality strategy reproduces much of the returns of the market while only being invested during the beginning and ends of the month and around holidays.

  52. DiggidyDan says:

    I know I.m late on this, but I have studied Mebane’s paper and thought the same thing, however, approached it in a different manner. My thoughts were this, use momentum indicators for the “signal, but on a conditional basis based on historical valuation comparison. . . . For instance, if the 200 day SMA crosses spy and CAPE is below 1 stdev off the average, buy. Provably too many permutations of that to backtest though

  53. Scott Teresi says:

    One aspect that I don’t believe Mebane’s paper tested at all is whether the 10-month moving average is robust for other days of the month, or if it’s just been lucky by picking the last day of the month. Would the system work nearly as well on day 10 of the month? Day 20? If there’s large variability, it might show the strategy will be brittle going forward.

    In effect, testing other days of the month would greatly expand the data available for backtesting.