A Review of Alpha Architect’s (Wes Gray/Jack Vogel) Quantitative Momentum book

This post will be an in-depth review of Alpha Architect’s Quantitative Momentum book. Overall, in my opinion, the book is terrific for those that are practitioners in fund management in the individual equity space, and still contains ideas worth thinking about outside of that space. However, the system detailed in the book benefits from nested ranking (rank along axis X, take the top decile, rank along axis Y within the top decile in X, and take the top decile along axis Y, essentially restricting selection to 1% of the universe). Furthermore, the book does not do much to touch upon volatility controls, which may have enhanced the system outlined greatly.

Before I get into the brunt of this post, I’d like to let my readers know that I formalized my nuts and bolts of quantstrat series of posts as a formal datacamp course. Datacamp is a very cheap way to learn a bunch of R, and financial applications are among those topics. My course covers the basics of quantstrat, and if those who complete the course like it, I may very well create more advanced quantstrat modules on datacamp. I’m hoping that the finance courses are well-received, since there are financial topics in R I’d like to learn myself that a 45 minute lecture doesn’t really suffice for (such as Dr. David Matteson’s change points magic, PortfolioAnalytics, and so on). In any case, here’s the link.

So, let’s start with a summary of the book:

Part 1 is several chapters that are the giant expose- of why momentum works (or at least, has worked for at least 20 years since 1993)…namely that human biases and irrational behaviors act in certain ways to make the anomaly work. Then there’s also the career risk (AKA it’s a risk factor, and so, if your benchmark is SPY and you run across a 3+ year period of underperformance, you have severe career risk), and essentially, a whole litany of why a professional asset manager would get fired but if you just stick with the anomaly over many many years and ride out multi-year stretches of relative underperformance, you’ll come out ahead in the very long run.

Generally, I feel like there’s work to be done if this is the best that can be done, but okay, I’ll accept it.

Essentially, part 1 is for the uninitiated. For those that have been around the momentum block a couple of times, they can skip right past this. Unfortunately, it’s half the book, so that leaves a little bit of a sour taste in the mouth.

Next, part two is where, in my opinion, the real meat and potatoes of the book–the “how”.

Essentially, the algorithm can be boiled down into the following:

Taking the universe of large and mid-cap stocks, do the following:

1) Sort the stocks into deciles by 2-12 momentum–that is, at the end of every month, calculate momentum by last month’s closing price minus the closing price 12 months ago. Essentially, research states that there’s a reversion effect on the 1-month momentum. However, this effect doesn’t carry over into the ETF universe in my experience.

2) Here’s the interesting part which makes the book worth picking up on its own (in my opinion): after sorting into deciles, rank the top decile by the following metric: multiply the sign of the 2-12 momentum by the following equation: (% negative returns – % positive). Essentially, the idea here is to determine smoothness of momentum. That is, in the most extreme situation, imagine a stock that did absolutely nothing for 230 days, and then had one massive day that gave it its entire price appreciation (think Google when it had a 10% jump off of better-than-expected numbers reports), and in the other extreme, a stock that simply had each and every single day be a small positive price appreciation. Obviously, you’d want the second type of stock. That’s this idea. Again, sort into deciles, and take the top decile. Therefore, taking the top decile of the top decile leaves you with 1% of the universe. Essentially, this makes the idea very difficult to replicate–since you’d need to track down a massive universe of stocks. That stated, I think the expression is actually a pretty good idea as a stand-in for volatility. That is, regardless of how volatile an asset is–whether it’s as volatile as a commodity like DBC, or as non-volatile as a fixed-income product like SHY, this expression is an interesting way of stating “this path is choppy” vs. “this path is smooth”. I might investigate this expression on my blog further in the future.

3) Lastly, if the portfolio is turning over quarterly instead of monthly, the best months to turn it over are the months preceding end-of-quarter month (that is, February, May, August, November) because a bunch of amateur asset managers like to “window dress” their portfolios. That is, they had a crummy quarter, so at the last month before they have to send out quarterly statements, they load up on some recent winners so that their clients don’t think they’re as amateur as they really let on, and there’s a bump for this. Similarly, January has some selling anomalies due to tax-loss harvesting. As far as practical implementations go, I think this is a very nice touch. Conceding the fact that turning over every month may be a bit too expensive, I like that Wes and Jack say “sure, you want to turn it over once every three months, but on *which* months?”. It’s a very good question to ask if it means you get an additional percentage point or 150 bps a year from that, as it just might cover the transaction costs and then some.

All in all, it’s a fairly simple to understand strategy. However, the part that sort of gates off the book to a perfect replication is the difficulty in obtaining the CRSP data.

However, I do commend Alpha Architect for disclosing the entire algorithm from start to finish.

Furthermore, if the basic 2-12 momentum is not enough, there’s an appendix detailing other types of momentum ideas (earnings momentum, ranking by distance to 52-week highs, absolute historical momentum, and so on). None of these strategies are really that much better than the basic price momentum strategy, so they’re there for those interested, but it seems there’s nothing really ground-breaking there. That is, if you’re trading once a month, there’s only so many ways of saying “hey, I think this thing is going up!”

I also like that Wes and Jack touched on the fact that trend-following, while it doesn’t improve overall CAGR or Sharpe, does a massive amount to improve on max drawdown. That is, if faced with the prospect of losing 70-80% of everything, and losing only 30%, that’s an easy choice to make. Trend-following is good, even a simplistic version.

All in all, I think the book accomplishes what it sets out to do, which is to present a well-researched algorithm. Ultimately, the punchline is on Alpha Architect’s site (I believe they have some sort of monthly stock filter). Furthermore, the book states that there are better risk-adjusted returns when combined with the algorithm outlined in the “quantitative value” book. In my experience, I’ve never had value algorithms impress me in the backtests I’ve done, but I can chalk that up to me being inexperienced with all the various valuation metrics.

My criticism of the book, however, is this:

The momentum algorithm in the book misses what I feel is one key component: volatility targeting control. Simply, the paper “momentum has its moments” (which I covered in my hypothesis-driven development series of posts) essentially states that the usual Fama-French momentum strategy does far better from a risk-reward strategy by deleveraging during times of excessive volatility, and avoiding momentum crashes. I’m not sure why Wes and Jack didn’t touch upon this paper, since the implementation is very simple (target/realized volatility = leverage factor). Ideally, I’d love if Wes or Jack could send me the stream of returns for this strategy (preferably daily, but monthly also works).

Essentially, I think this book is very comprehensive. However, I think it also has a somewhat “don’t try this at home” feel to it due to the data requirement to replicate it. Certainly, if your broker charges you $8 a transaction, it’s not a feasible strategy to drop several thousand bucks a year on transaction costs that’ll just give your returns to your broker. However, I do wonder if the QMOM ETF (from Alpha Architect, of course) is, in fact, a better version of this strategy, outside of the management fee.

In any case, my final opinion is this: while this book leaves a little bit of knowledge on the table, on a whole, it accomplishes what it sets out to do, is clear with its procedures, and provides several worthwhile ideas. For the price of a non-technical textbook (aka those $60+ books on amazon), this book is a steal.

4.5/5 stars.

Thanks for reading.

NOTE: While I am currently employed in a successful analytics capacity, I am interested in hearing about full-time positions more closely related to the topics on this blog. If you have a full-time position which can benefit from my current skills, please let me know. My Linkedin can be found here.

The Problem With Depmix For Online Regime Prediction

This post will be about attempting to use the Depmix package for online state prediction. While the depmix package performs admirably when it comes to describing the states of the past, when used for one-step-ahead prediction, under the assumption that tomorrow’s state will be identical to today’s, the hidden markov model process found within the package does not perform to expectations.

So, to start off, this post was motivated by Michael Halls-Moore, who recently posted some R code about using the depmixS4 library to use hidden markov models. Generally, I am loath to create posts on topics I don’t feel I have an absolutely front-to-back understanding of, but I’m doing this in the hope of learning from others on how to appropriately do online state-space prediction, or “regime switching” detection, as it may be called in more financial parlance.

Here’s Dr. Halls-Moore’s post.

While I’ve seen the usual theory of hidden markov models (that is, it can rain or it can be sunny, but you can only infer the weather judging by the clothes you see people wearing outside your window when you wake up), and have worked with toy examples in MOOCs (Udacity’s self-driving car course deals with them, if I recall correctly–or maybe it was the AI course), at the end of the day, theory is only as good as how well an implementation can work on real data.

For this experiment, I decided to take SPY data since inception, and do a full in-sample “backtest” on the data. That is, given that the HMM algorithm from depmix sees the whole history of returns, with this “god’s eye” view of the data, does the algorithm correctly classify the regimes, if the backtest results are any indication?

Here’s the code to do so, inspired by Dr. Halls-Moore’s.

require(depmixS4)
require(quantmod)
getSymbols('SPY', from = '1990-01-01', src='yahoo', adjust = TRUE)
spyRets <- na.omit(Return.calculate(Ad(SPY)))

set.seed(123)

hmm <- depmix(SPY.Adjusted ~ 1, family = gaussian(), nstates = 3, data=spyRets)
hmmfit <- fit(hmm, verbose = FALSE)
post_probs <- posterior(hmmfit)
post_probs <- xts(post_probs, order.by=index(spyRets))
plot(post_probs$state)
summaryMat <- data.frame(summary(hmmfit))
colnames(summaryMat) <- c("Intercept", "SD")
bullState <- which(summaryMat$Intercept > 0)
bearState <- which(summaryMat$Intercept < 0)

hmmRets <- spyRets * lag(post_probs$state == bullState) - spyRets * lag(post_probs$state == bearState)
charts.PerformanceSummary(hmmRets)
table.AnnualizedReturns(hmmRets)

Essentially, while I did select three states, I noted that anything with an intercept above zero is a bull state, and below zero is a bear state, so essentially, it reduces to two states.

With the result:

table.AnnualizedReturns(hmmRets)
                          SPY.Adjusted
Annualized Return               0.1355
Annualized Std Dev              0.1434
Annualized Sharpe (Rf=0%)       0.9448

So, not particularly terrible. The algorithm works, kind of, sort of, right?

Well, let’s try online prediction now.

require(DoMC)

dailyHMM <- function(data, nPoints) {
  subRets <- data[1:nPoints,]
  hmm <- depmix(SPY.Adjusted ~ 1, family = gaussian(), nstates = 3, data = subRets)
  hmmfit <- fit(hmm, verbose = FALSE)
  post_probs <- posterior(hmmfit)
  summaryMat <- data.frame(summary(hmmfit))
  colnames(summaryMat) <- c("Intercept", "SD")
  bullState <- which(summaryMat$Intercept > 0)
  bearState <- which(summaryMat$Intercept < 0)
  if(last(post_probs$state) %in% bullState) {
    state <- xts(1, order.by=last(index(subRets)))
  } else if (last(post_probs$state) %in% bearState) {
    state <- xts(-1, order.by=last(index(subRets)))
  } else {
    state <- xts(0, order.by=last(index(subRets)))
  }
  colnames(state) <- "State"
  return(state)
}

# took 3 hours in parallel
t1 <- Sys.time()
set.seed(123)
registerDoMC((detectCores() - 1))
states <- foreach(i = 500:nrow(spyRets), .combine=rbind) %dopar% {
  dailyHMM(data = spyRets, nPoints = i)
}
t2 <- Sys.time()
print(t2-t1)

So what I did here was I took an expanding window, starting from 500 days since SPY’s inception, and kept increasing it, by one day at a time. My prediction, was, trivially enough, the most recent day, using a 1 for a bull state, and a -1 for a bear state. I ran this process in parallel (on a linux cluster, because windows’s doParallel library seems to not even know that certain packages are loaded, and it’s more messy), and the first big issue is that this process took about three hours on seven cores for about 23 years of data. Not exactly encouraging, but computing time isn’t expensive these days.

So let’s see if this process actually works.

First, let’s test if the algorithm does what it’s actually supposed to do and use one day of look-ahead bias (that is, the algorithm tells us the state at the end of the day–how correct is it even for that day?).


onlineRets <- spyRets * states 
charts.PerformanceSummary(onlineRets)
table.AnnualizedReturns(onlineRets)

With the result:

> table.AnnualizedReturns(onlineRets)
                          SPY.Adjusted
Annualized Return               0.2216
Annualized Std Dev              0.1934
Annualized Sharpe (Rf=0%)       1.1456

So, allegedly, the algorithm seems to do what it was designed to do, which is to classify a state for a given data set. Now, the most pertinent question: how well do these predictions do even one day ahead? You’d think that state space predictions would be parsimonious from day to day, given the long history, correct?


onlineRets <- spyRets * lag(states)
charts.PerformanceSummary(onlineRets)
table.AnnualizedReturns(onlineRets)

With the result:

> table.AnnualizedReturns(onlineRets)
                          SPY.Adjusted
Annualized Return               0.0172
Annualized Std Dev              0.1939
Annualized Sharpe (Rf=0%)       0.0888

That is, without the lookahead bias, the state space prediction algorithm is atrocious. Why is that?

Well, here’s the plot of the states:

In short, the online hmm algorithm in the depmix package seems to change its mind very easily, with obvious (negative) implications for actual trading strategies.

So, that wraps it up for this post. Essentially, the main message here is this: there’s a vast difference between loading doing descriptive analysis (AKA “where have you been, why did things happen”) vs. predictive analysis (that is, “if I correctly predict the future, I get a positive payoff”). In my opinion, while descriptive statistics have their purpose in terms of explaining why a strategy may have performed how it did, ultimately, we’re always looking for better prediction tools. In this case, depmix, at least in this “out-of-the-box” demonstration does not seem to be the tool for that.

If anyone has had success with using depmix (or other regime-switching algorithm in R) for prediction, I would love to see work that details the procedure taken, as it’s an area I’m looking to expand my toolbox into, but don’t have any particular good leads. Essentially, I’d like to think of this post as me describing my own experiences with the package.

Thanks for reading.

NOTE: On Oct. 5th, I will be in New York City. On Oct. 6th, I will be presenting at The Trading Show on the Programming Wars panel.

NOTE: My current analytics contract is up for review at the end of the year, so I am officially looking for other offers as well. If you have a full-time role which may benefit from the skills you see on my blog, please get in touch with me. My linkedin profile can be found here.

An Introduction to Portfolio Component Conditional Value At Risk

This post will introduce component conditional value at risk mechanics found in PerformanceAnalytics from a paper written by Brian Peterson, Kris Boudt, and Peter Carl. This is a mechanism that is an easy-to-call mechanism for computing component expected shortfall in asset returns as they apply to a portfolio. While the exact mechanics are fairly complex, the upside is that the running time is nearly instantaneous, and this method is a solid tool for including in asset allocation analysis.

For those interested in an in-depth analysis of the intuition of component conditional value at risk, I refer them to the paper written by Brian Peterson, Peter Carl, and Kris Boudt.

Essentially, here’s the idea: all assets in a given portfolio have a marginal contribution to its total conditional value at risk (also known as expected shortfall)–that is, the expected loss when the loss surpasses a certain threshold. For instance, if you want to know your 5% expected shortfall, then it’s the average of the worst 5 returns per 100 days, and so on. For returns using daily resolution, the idea of expected shortfall may sound as though there will never be enough data in a sufficiently fast time frame (on one year or less), the formula for expected shortfall in the PerformanceAnalytics defaults to an approximation calculation using a Cornish-Fisher expansion, which delivers very good results so long as the p-value isn’t too extreme (that is, it works for relatively sane p values such as the 1%-10% range).

Component Conditional Value at Risk has two uses: first off, given no input weights, it uses an equal weight default, which allows it to provide a risk estimate for each individual asset without burdening the researcher to create his or her own correlation/covariance heuristics. Secondly, when provided with a set of weights, the output changes to reflect the contribution of various assets in proportion to those weights. This means that this methodology works very nicely with strategies that exclude assets based on momentum, but need a weighting scheme for the remaining assets. Furthermore, using this methodology also allows an ex-post analysis of risk contribution to see which instrument contributed what to risk.

First, a demonstration of how the mechanism works using the edhec data set. There is no strategy here, just a demonstration of syntax.

require(quantmod)
require(PerformanceAnalytics)

data(edhec)

tmp &lt;- CVaR(edhec, portfolio_method = &quot;component&quot;)

This will assume an equal-weight contribution from all of the funds in the edhec data set.

So tmp is the contribution to expected shortfall from each of the various edhec managers over the entire time period. Here’s the output:

$MES
           [,1]
[1,] 0.03241585

$contribution
 Convertible Arbitrage             CTA Global  Distressed Securities       Emerging Markets  Equity Market Neutral
          0.0074750513          -0.0028125166           0.0039422674           0.0069376579           0.0008077760
          Event Driven Fixed Income Arbitrage           Global Macro      Long/Short Equity       Merger Arbitrage
          0.0037114666           0.0043125937           0.0007173036           0.0036152960           0.0013693293
        Relative Value          Short Selling         Funds of Funds
          0.0037650911          -0.0048178690           0.0033924063 

$pct_contrib_MES
 Convertible Arbitrage             CTA Global  Distressed Securities       Emerging Markets  Equity Market Neutral
            0.23059863            -0.08676361             0.12161541             0.21402052             0.02491917
          Event Driven Fixed Income Arbitrage           Global Macro      Long/Short Equity       Merger Arbitrage
            0.11449542             0.13303965             0.02212817             0.11152864             0.04224258
        Relative Value          Short Selling         Funds of Funds
            0.11614968            -0.14862694             0.10465269

The salient part of this is the percent contribution (the last output). Notice that it can be negative, meaning that certain funds gain when others lose. At least, this was the case over the current data set. These assets diversify a portfolio and actually lower expected shortfall.

&gt; tmp2 &lt;- CVaR(edhec, portfolio_method = &quot;component&quot;, weights = c(rep(.1, 10), rep(0,3)))
&gt; tmp2
$MES
           [,1]
[1,] 0.04017453

$contribution
 Convertible Arbitrage             CTA Global  Distressed Securities       Emerging Markets  Equity Market Neutral
          0.0086198045          -0.0046696862           0.0058778855           0.0109152240           0.0009596620
          Event Driven Fixed Income Arbitrage           Global Macro      Long/Short Equity       Merger Arbitrage
          0.0054824325           0.0050398011           0.0009638502           0.0044568333           0.0025287234
        Relative Value          Short Selling         Funds of Funds
          0.0000000000           0.0000000000           0.0000000000 

$pct_contrib_MES
 Convertible Arbitrage             CTA Global  Distressed Securities       Emerging Markets  Equity Market Neutral
            0.21455894            -0.11623499             0.14630875             0.27169512             0.02388732
          Event Driven Fixed Income Arbitrage           Global Macro      Long/Short Equity       Merger Arbitrage
            0.13646538             0.12544767             0.02399157             0.11093679             0.06294345
        Relative Value          Short Selling         Funds of Funds
            0.00000000             0.00000000             0.00000000

In this case, I equally weighted the first ten managers in the edhec data set, and put zero weight in the last three. Furthermore, we can see what happens when the weights are not equal.

&gt; tmp3 &lt;- CVaR(edhec, portfolio_method = &quot;component&quot;, weights = c(.2, rep(.1, 9), rep(0,3)))
&gt; tmp3
$MES
           [,1]
[1,] 0.04920372

$contribution
 Convertible Arbitrage             CTA Global  Distressed Securities       Emerging Markets  Equity Market Neutral
          0.0187406982          -0.0044391078           0.0057235762           0.0102706768           0.0007710434
          Event Driven Fixed Income Arbitrage           Global Macro      Long/Short Equity       Merger Arbitrage
          0.0051541429           0.0055944367           0.0008028457           0.0044085104           0.0021768951
        Relative Value          Short Selling         Funds of Funds
          0.0000000000           0.0000000000           0.0000000000 

$pct_contrib_MES
 Convertible Arbitrage             CTA Global  Distressed Securities       Emerging Markets  Equity Market Neutral
            0.38087972            -0.09021895             0.11632406             0.20873782             0.01567043
          Event Driven Fixed Income Arbitrage           Global Macro      Long/Short Equity       Merger Arbitrage
            0.10475109             0.11369947             0.01631677             0.08959710             0.04424249
        Relative Value          Short Selling         Funds of Funds
            0.00000000             0.00000000             0.00000000

This time, notice that as the weight increased in the convertible arb manager, so too did his contribution to maximum expected shortfall.

For a future backtest, I would like to make some data requests. I would like to use the universe found in Faber’s Global Asset Allocation book. That said, the simulations in that book go back to 1972, and I was wondering if anyone out there has daily returns for those assets/indices. While some ETFs go back into the early 2000s, there are some that start rather late such as DBC (commodities, early 2006), GLD (gold, early 2004), BWX (foreign bonds, late 2007), and FTY (NAREIT, early 2007). As an eight-year backtest would be a bit short, I was wondering if anyone had data with more history.

One other thing, I will in New York for the trading show, and speaking on the “programming wars” panel on October 6th.

Thanks for reading.

NOTE: While I am currently contracting, I am also looking for a permanent position which can benefit from my skills for when my current contract ends. If you have or are aware of such an opening, I will be happy to speak with you.

A Return.Portfolio Wrapper to Automate Harry Long Seeking Alpha Backtests

This post will cover a function to simplify creating Harry Long type rebalancing strategies from SeekingAlpha for interested readers. As Harry Long has stated, most, if not all of his strategies are more for demonstrative purposes rather than actual recommended investments.

So, since Harry Long has been posting some more articles on Seeknig Alpha, I’ve had a reader or two ask me to analyze his strategies (again). Instead of doing that, however, I’ll simply put this tool here, which is a wrapper that automates the acquisition of data and simulates portfolio rebalancing with one line of code.

Here’s the tool.

require(quantmod)
require(PerformanceAnalytics)
require(downloader)

LongSeeker <- function(symbols, weights, rebalance_on = "years", 
                       displayStats = TRUE, outputReturns = FALSE) {
  getSymbols(symbols, src='yahoo', from = '1990-01-01')
  prices <- list()
  for(i in 1:length(symbols)) {
    if(symbols[i] == "ZIV") {
      download("https://www.dropbox.com/s/jk3ortdyru4sg4n/ZIVlong.TXT", destfile="ziv.txt")
      ziv <- xts(read.zoo("ziv.txt", header=TRUE, sep=",", format="%Y-%m-%d"))
      prices[[i]] <- Cl(ziv)
    } else if (symbols[i] == "VXX") {
      download("https://dl.dropboxusercontent.com/s/950x55x7jtm9x2q/VXXlong.TXT", 
               destfile="vxx.txt")
      vxx <- xts(read.zoo("vxx.txt", header=TRUE, sep=",", format="%Y-%m-%d"))
      prices[[i]] <- Cl(vxx)
    }
    else {
      prices[[i]] <- Ad(get(symbols[i]))
    }
  }
  prices <- do.call(cbind, prices)
  prices <- na.locf(prices)
  returns <- na.omit(Return.calculate(prices))
  
  returns$zeroes <- 0
  weights <- c(weights, 1-sum(weights))
  stratReturns <- Return.portfolio(R = returns, weights = weights, rebalance_on = rebalance_on)
  
  if(displayStats) {
    stats <- rbind(table.AnnualizedReturns(stratReturns), maxDrawdown(stratReturns), CalmarRatio(stratReturns))
    rownames(stats)[4] <- "Max Drawdown"
    print(stats)
    charts.PerformanceSummary(stratReturns)
  }
  
  if(outputReturns) {
    return(stratReturns)
  }
} 

It fetches the data for you (usually from Yahoo, but a big thank you to Mr. Helumth Vollmeier in the case of ZIV and VXX), and has the option of either simply displaying an equity curve and some statistics (CAGR, annualized standard dev, Sharpe, max Drawdown, Calmar), or giving you the return stream as an output if you wish to do more analysis in R.

Here’s an example of simply getting the statistics, with an 80% XLP/SPLV (they’re more or less interchangeable) and 20% TMF (aka 60% TLT, so an 80/60 portfolio), from one of Harry Long’s articles.

LongSeeker(c("XLP", "TLT"), c(.8, .6))

Statistics:


                          portfolio.returns
Annualized Return                 0.1321000
Annualized Std Dev                0.1122000
Annualized Sharpe (Rf=0%)         1.1782000
Max Drawdown                      0.2330366
Calmar Ratio                      0.5670285

Equity curve:

Nothing out of the ordinary of what we might expect from a balanced equity/bonds portfolio. Generally does well, has its largest drawdown in the financial crisis, and some other bumps in the road, but overall, I’d think a fairly vanilla “set it and forget it” sort of thing.

And here would be the way to get the stream of individual daily returns, assuming you wanted to rebalance these two instruments weekly, instead of yearly (as is the default).

tmp <- LongSeeker(c("XLP", "TLT"), c(.8, .6), rebalance_on="weeks",
                    displayStats = FALSE, outputReturns = TRUE)

And now let’s get some statistics.

table.AnnualizedReturns(tmp)
maxDrawdown(tmp)
CalmarRatio(tmp)

Which give:

> table.AnnualizedReturns(tmp)
                          portfolio.returns
Annualized Return                    0.1328
Annualized Std Dev                   0.1137
Annualized Sharpe (Rf=0%)            1.1681
> maxDrawdown(tmp)
[1] 0.2216417
> CalmarRatio(tmp)
             portfolio.returns
Calmar Ratio         0.5990087

Turns out, moving the rebalancing from annually to weekly didn’t have much of an effect here (besides give a bunch of money to your broker, if you factored in transaction costs, which this doesn’t).

So, that’s how this tool works. The results, of course, begin from the latest instrument’s inception. The trick, in my opinion, is to try and find proxy substitutes with longer histories for newer ETFs that are simply leveraged ETFs, such as using a 60% weight in TLT with an 80% weight in XLP instead of a 20% weight in TMF with 80% allocation in SPLV.

For instance, here are some proxies:

SPXL = XLP
SPXL/UPRO = SPY * 3
TMF = TLT * 3

That said, I’ve worked with Harry Long before, and he develops more sophisticated strategies behind the scenes, so I’d recommend that SeekingAlpha readers take his publicly released strategies as concept demonstrations, as opposed to fully-fledged investment ideas, and contact Mr. Long himself about more customized, private solutions for investment institutions if you are so interested.

Thanks for reading.

NOTE: I am currently in the northeast. While I am currently contracting, I am interested in networking with individuals or firms with regards to potential collaboration opportunities.

How To Compute Turnover With Return.Portfolio in R

This post will demonstrate how to take into account turnover when dealing with returns-based data using PerformanceAnalytics and the Return.Portfolio function in R. It will demonstrate this on a basic strategy on the nine sector SPDRs.

So, first off, this is in response to a question posed by one Robert Wages on the R-SIG-Finance mailing list. While there are many individuals out there with a plethora of questions (many of which can be found to be demonstrated on this blog already), occasionally, there will be an industry veteran, a PhD statistics student from Stanford, or other very intelligent individual that will ask a question on a topic that I haven’t yet touched on this blog, which will prompt a post to demonstrate another technical aspect found in R. This is one of those times.

So, this demonstration will be about computing turnover in returns space using the PerformanceAnalytics package. Simply, outside of the PortfolioAnalytics package, PerformanceAnalytics with its Return.Portfolio function is the go-to R package for portfolio management simulations, as it can take a set of weights, a set of returns, and generate a set of portfolio returns for analysis with the rest of PerformanceAnalytics’s functions.

Again, the strategy is this: take the 9 three-letter sector SPDRs (since there are four-letter ETFs now), and at the end of every month, if the adjusted price is above its 200-day moving average, invest into it. Normalize across all invested sectors (that is, 1/9th if invested into all 9, 100% into 1 if only 1 invested into, 100% cash, denoted with a zero return vector, if no sectors are invested into). It’s a simple, toy strategy, as the strategy isn’t the point of the demonstration.

Here’s the basic setup code:

require(TTR)
require(PerformanceAnalytics)
require(quantmod)

symbols <- c("XLF", "XLK", "XLU", "XLE", "XLP", "XLF", "XLB", "XLV", "XLY")
getSymbols(symbols, src='yahoo', from = '1990-01-01-01')
prices <- list()
for(i in 1:length(symbols)) {
  tmp <- Ad(get(symbols[[i]]))
  prices[[i]] <- tmp
}
prices <- do.call(cbind, prices)

# Our signal is a simple adjusted price over 200 day SMA
signal <- prices > xts(apply(prices, 2, SMA, n = 200), order.by=index(prices))

# equal weight all assets with price above SMA200
returns <- Return.calculate(prices)
weights <- signal/(rowSums(signal)+1e-16)

# With Return.portfolio, need all weights to sum to 1
weights$zeroes <- 1 - rowSums(weights)
returns$zeroes <- 0

monthlyWeights <- na.omit(weights[endpoints(weights, on = 'months'),])
weights <- na.omit(weights)
returns <- na.omit(returns)

So, get the SPDRs, put them together, compute their returns, generate the signal, and create the zero vector, since Return.Portfolio treats weights less than 1 as a withdrawal, and weights above 1 as the addition of more capital (big FYI here).

Now, here’s how to compute turnover:

out <- Return.portfolio(R = returns, weights = monthlyWeights, verbose = TRUE)
beginWeights <- out$BOP.Weight
endWeights <- out$EOP.Weight
txns <- beginWeights - lag(endWeights)
monthlyTO <- xts(rowSums(abs(txns[,1:9])), order.by=index(txns))
plot(monthlyTO)

So, the trick is this: when you call Return.portfolio, use the verbose = TRUE option. This creates several objects, among them returns, BOP.Weight, and EOP.Weight. These stand for Beginning Of Period Weight, and End Of Period Weight.

The way that turnover is computed is simply the difference between how the day’s return moves the allocated portfolio from its previous ending point to where that portfolio actually stands at the beginning of next period. That is, the end of period weight is the beginning of period drift after taking into account the day’s drift/return for that asset. The new beginning of period weight is the end of period weight plus any transacting that would have been done. Thus, in order to find the actual transactions (or turnover), one subtracts the previous end of period weight from the beginning of period weight.

This is what such transactions look like for this strategy.

Something we can do with such data is take a one-year rolling turnover, accomplished with the following code:

yearlyTO <- runSum(monthlyTO, 252)
plot(yearlyTO, main = "running one year turnover")

It looks like this:

This essentially means that one year’s worth of two-way turnover (that is, if selling an entirely invested portfolio is 100% turnover, and buying an entirely new set of assets is another 100%, then two-way turnover is 200%) is around 800% at maximum. That may be pretty high for some people.

Now, here’s the application when you penalize transaction costs at 20 basis points per percentage point traded (that is, it costs 20 cents to transact $100).

txnCosts <- monthlyTO * -.0020
retsWithTxnCosts <- out$returns + txnCosts
compare <- na.omit(cbind(out$returns, retsWithTxnCosts))
colnames(compare) <- c("NoTxnCosts", "TxnCosts20BPs")
charts.PerformanceSummary(compare)
table.AnnualizedReturns(compare)

And the result:


                          NoTxnCosts TxnCosts20BPs
Annualized Return             0.0587        0.0489
Annualized Std Dev            0.1554        0.1553
Annualized Sharpe (Rf=0%)     0.3781        0.3149

So, at 20 basis points on transaction costs, that takes about one percent in returns per year out of this (admittedly, terrible) strategy. This is far from negligible.

So, that is how you actually compute turnover and transaction costs. In this case, the transaction cost model was very simple. However, given that Return.portfolio returns transactions at the individual asset level, one could get as complex as they would like with modeling the transaction costs.

Thanks for reading.

NOTE: I will be giving a lightning talk at R/Finance, so for those attending, you’ll be able to find me there.

Create Amazing Looking Backtests With This One Wrong–I Mean Weird–Trick! (And Some Troubling Logical Invest Results)

This post will outline an easy-to-make mistake in writing vectorized backtests–namely in using a signal obtained at the end of a period to enter (or exit) a position in that same period. The difference in results one obtains is massive.

Today, I saw two separate posts from Alpha Architect and Mike Harris both referencing a paper by Valeriy Zakamulin on the fact that some previous trend-following research by Glabadanidis was done with shoddy results, and that Glabadanidis’s results were only reproducible through instituting lookahead bias.

The following code shows how to reproduce this lookahead bias.

First, the setup of a basic moving average strategy on the S&P 500 index from as far back as Yahoo data will provide.

require(quantmod)
require(xts)
require(TTR)
require(PerformanceAnalytics)

getSymbols('^GSPC', src='yahoo', from = '1900-01-01')
monthlyGSPC <- Ad(GSPC)[endpoints(GSPC, on = 'months')]

# change this line for signal lookback
movAvg <- SMA(monthlyGSPC, 10)

signal <- monthlyGSPC > movAvg
gspcRets <- Return.calculate(monthlyGSPC)

And here is how to institute the lookahead bias.

lookahead <- signal * gspcRets
correct <- lag(signal) * gspcRets

These are the “results”:

compare <- na.omit(cbind(gspcRets, lookahead, correct))
colnames(compare) <- c("S&P 500", "Lookahead", "Correct")
charts.PerformanceSummary(compare)
rbind(table.AnnualizedReturns(compare), maxDrawdown(compare), CalmarRatio(compare))
logRets <- log(cumprod(1+compare))
chart.TimeSeries(logRets, legend.loc='topleft')

Of course, this equity curve is of no use, so here’s one in log scale.

As can be seen, lookahead bias makes a massive difference.

Here are the numerical results:

                            S&P 500  Lookahead   Correct
Annualized Return         0.0740000 0.15550000 0.0695000
Annualized Std Dev        0.1441000 0.09800000 0.1050000
Annualized Sharpe (Rf=0%) 0.5133000 1.58670000 0.6623000
Worst Drawdown            0.5255586 0.08729914 0.2699789
Calmar Ratio              0.1407286 1.78119192 0.2575219

Again, absolutely ridiculous.

Note that when using Return.Portfolio (the function in PerformanceAnalytics), that package will automatically give you the next period’s return, instead of the current one, for your weights. However, for those writing “simple” backtests that can be quickly done using vectorized operations, an off-by-one error can make all the difference between a backtest in the realm of reasonable, and pure nonsense. However, should one wish to test for said nonsense when faced with impossible-to-replicate results, the mechanics demonstrated above are the way to do it.

Now, onto other news: I’d like to thank Gerald M for staying on top of one of the Logical Invest strategies–namely, their simple global market rotation strategy outlined in an article from an earlier blog post.

Up until March 2015 (the date of the blog post), the strategy had performed well. However, after said date?

It has been a complete disaster, which, in hindsight, was evident when I passed it through the hypothesis-driven development framework process I wrote about earlier.

So, while there has been a great deal written about not simply throwing away a strategy because of short-term underperformance, and that anomalies such as momentum and value exist because of career risk due to said short-term underperformance, it’s never a good thing when a strategy creates historically large losses, particularly after being published in such a humble corner of the quantitative financial world.

In any case, this was a post demonstrating some mechanics, and an update on a strategy I blogged about not too long ago.

Thanks for reading.

NOTE: I am always interested in hearing about new opportunities which may benefit from my expertise, and am always happy to network. You can find my LinkedIn profile here.

Are R^2s Useful In Finance? Hypothesis-Driven Development In Reverse

This post will shed light on the values of R^2s behind two rather simplistic strategies — the simple 10 month SMA, and its relative, the 10 month momentum (which is simply a difference of SMAs, as Alpha Architect showed in their book DIY Financial Advisor.

Not too long ago, a friend of mine named Josh asked me a question regarding R^2s in finance. He’s finishing up his PhD in statistics at Stanford, so when people like that ask me questions, I’d like to answer them. His assertion is that in some instances, models that have less than perfect predictive power (EG R^2s of .4, for instance), can still deliver very promising predictions, and that if someone were to have a financial model that was able to explain 40% of the variance of returns, they could happily retire with that model making them very wealthy. Indeed, .4 is a very optimistic outlook (to put it lightly), as this post will show.

In order to illustrate this example, I took two “staple” strategies — buy SPY when its closing monthly price is above its ten month simple moving average, and when its ten month momentum (basically the difference of a ten month moving average and its lag) is positive. While these models are simplistic, they are ubiquitously talked about, and many momentum strategies are an improvement upon these baseline, “out-of-the-box” strategies.

Here’s the code to do that:

require(xts)
require(quantmod)
require(PerformanceAnalytics)
require(TTR)

getSymbols('SPY', from = '1990-01-01', src = 'yahoo')
adjustedPrices <- Ad(SPY)
monthlyAdj <- to.monthly(adjustedPrices, OHLC=TRUE)

spySMA <- SMA(Cl(monthlyAdj), 10)
spyROC <- ROC(Cl(monthlyAdj), 10)
spyRets <- Return.calculate(Cl(monthlyAdj))

smaRatio <- Cl(monthlyAdj)/spySMA - 1
smaSig <- smaRatio > 0
rocSig <- spyROC > 0

smaRets <- lag(smaSig) * spyRets
rocRets <- lag(rocSig) * spyRets

And here are the results:

strats <- na.omit(cbind(smaRets, rocRets, spyRets))
colnames(strats) <- c("SMA10", "MOM10", "BuyHold")
charts.PerformanceSummary(strats, main = "strategies")
rbind(table.AnnualizedReturns(strats), maxDrawdown(strats), CalmarRatio(strats))

                              SMA10     MOM10   BuyHold
Annualized Return         0.0975000 0.1039000 0.0893000
Annualized Std Dev        0.1043000 0.1080000 0.1479000
Annualized Sharpe (Rf=0%) 0.9346000 0.9616000 0.6035000
Worst Drawdown            0.1663487 0.1656176 0.5078482
Calmar Ratio              0.5860332 0.6270657 0.1757849

In short, the SMA10 and the 10-month momentum (aka ROC 10 aka MOM10) both handily outperform the buy and hold, not only in absolute returns, but especially in risk-adjusted returns (Sharpe and Calmar ratios). Again, simplistic analysis, and many models get much more sophisticated than this, but once again, simple, illustrative example using two strategies that outperform a benchmark (over the long term, anyway).

Now, the question is, what was the R^2 of these models? To answer this, I took a rolling five-year window that essentially asked: how well did these quantities (the ratio between the closing price and the moving average – 1, or the ten month momentum) predict the next month’s returns? That is, what proportion of the variance is explained through the monthly returns regressed against the previous month’s signals in numerical form (perhaps not the best framing, as the signal is binary as opposed to continuous which is what is being regressed, but let’s set that aside, again, for the sake of illustration).

Here’s the code to generate the answer.

predictorsAndPredicted <- na.omit(cbind(lag(smaRatio), lag(spyROC), spyRets))
R2s <- list()
for(i in 1:(nrow(predictorsAndPredicted)-59))  { #rolling five-year regression
  subset <- predictorsAndPredicted[i:(i+59),]
  smaLM <- lm(subset[,3]~subset[,1])
  smaR2 <- summary(smaLM)$r.squared
  rocLM <- lm(subset[,3]~subset[,2])
  rocR2 <- summary(rocLM)$r.squared
  R2row <- xts(cbind(smaR2, rocR2), order.by=last(index(subset)))
  R2s[[i]] <- R2row
}
R2s <- do.call(rbind, R2s)
par(mfrow=c(1,1))
colnames(R2s) <- c("SMA", "Momentum")
chart.TimeSeries(R2s, main = "R2s", legend.loc = 'topleft')

And the answer, in pictorial form:

In short, even in the best case scenarios, namely, crises which provide momentum/trend-following/call it what you will its raison d’etre, that is, its risk management appeal, the proportion of variance explained by the actual signal quantities was very small. In the best of times, around 20%. But then again, think about what the R^2 value actually is–it’s the percentage of variance explained by a predictor. If a small set of signals (let alone one) was able to explain the majority of the change in the returns of the S&P 500, or even a not-insignificant portion, such a person would stand to become very wealthy. More to the point, given that two strategies that handily outperform the market have R^2s that are exceptionally low for extended periods of time, it goes to show that holding the R^2 up as some form of statistical holy grail certainly is incorrect in the general sense, and anyone who does so either is painting with too broad a brush, is creating disingenuous arguments, or should simply attempt to understand another field which may not work the way their intuition tells them.

Thanks for reading.