Tampilkan postingan dengan label factor model. Tampilkan semua postingan
Tampilkan postingan dengan label factor model. Tampilkan semua postingan

Senin, 27 Juli 2020

Do Element Models Live On Inwards The Curt Term?

Besides pair-trading, “factor model” is the close pop workhorse of the statistical arbitrageur. In a previous article, I discussed the close well-known ingredient model – the Fama-French Three-Factor model, alongside the full general marketplace index returns, the market-cap of the stock, too the book-to-price ratio equally the exclusively iii factors driving returns. However, equally I explained earlier, this ingredient model has a really long horizon. For the quantitative trader who needs to brand coin every month, the natural instinct is to expect for a to a greater extent than “sophisticated” ingredient that plant inwards the brusk term, or fifty-fifty to railroad train roughly variety of model that role dissimilar factors every calendar month inwards answer to “market condition”. Alas, other than hearsays too second-hand gossips, I accept never witnessed an actual success of this approach inwards a hedge fund or proprietary trading grouping – at to the lowest degree a success that lasts for to a greater extent than than a year.

I am of course of didactics non privy to the electrical flow surgical procedure numbers of ingredient models run past times roughly of the close successful hedge funds today. However, in that location is a shape of ETF (called “XTF”) marketed past times PowerShares Capital Management that uses a similar ingredient approach for its stock pick criteria. According to media reports, each stock inwards these XTF’s is scored past times 25 variables such equally cash flow, earnings growth, cost momentum, etc. This sounds similar a classic ingredient model to me. This model is reportedly designed past times the quantitative unit of measurement at American Stock Exchange. To abide by out if they accept indeed discovered the holy grail of ingredient models, I looked at the surgical procedure of these XTF compared to their benchmarks.

Here I tabulate the XTF’s for each marketplace cap too value category, their corresponding benchmark marketplace index ETF’s, too finally the YTD differential returns upwards to Dec 13, 2006. (PJG too PJM accept equally good brusk a history for this comparison.)










ValueBlendGrowth
Large capPWV-IVE=4.8%PWC-IVV=-3.6%PWB-IVW=-5.0%
Mid capPWP-IJJ=0.1%PJG-IJH=N/APWJ-IJK=3.1%
Small capPWY-IJS=-0.7%PJM-IJR=N/APWT-IJT=-4.9%




The differential returns are all over the place: roughly positive, others negative. To me, this is symptomatic of a ingredient model that does non accept predictive power. (After all, if the differential returns are consistently negative, nosotros could accept long the ETF, brusk the XTF, too brand consistent profits!) At the really least, this ingredient model may accept a horizon much longer than what close traders would live interested inwards – inwards which case, why non but role the uncomplicated Fama-French model?

This is non to tell that exotic, proprietary ingredient models accept no use: they tend to live pretty useful for peril management, equally volatilities too correlations are ofttimes easier to predict than returns. But beware every fourth dimension your peril administration software vendor tries to sell yous an alpha generator!

Senin, 13 Juli 2020

Seven Factors That Capture Virtually Of Hedge Funds' Returns

The Economist magazine simply published an article on this topic previously.) The 7 factors are (are you lot ready?):

1) excess provide on the S&P 500 index;
2) a pocket-size minus large part constructed every bit the deviation of the Wilshire pocket-size too large
capitalization stock indices;
3) excess returns on portfolios of lookback straddle options on currencies;
4) excess returns on portfolios of lookback straddle options on commodities;
5) excess returns on portfolios of lookback straddle options on bonds;
6) the yield spread of the United States of America 10 twelvemonth treasury bond over the iii calendar month T-bill, adjusted for the duration of the 10 twelvemonth bond;
7) the modify inwards the credit spread of the Moody's BAA bond over the 10 twelvemonth treasury bond, too appropriately adjusted for duration.

According to the researchers, factors 3)-5) are constructed to replicate the maximum possible provide to trend-following strategies on their respective underlying assets.

See, it is non that hard to run a hedge fund later all!

Senin, 06 Juli 2020

A Component Subdivision Model That I Tin Order The Sack Believe In

Some of yous may retrieve that I preached about the uselessness of constituent models inwards predicting curt term return, together with the unreliability of many exotic factors fifty-fifty for the long term. In particular, constituent models are particularly inaccurate inwards valuing increase stocks (i.e. stocks amongst depression book-to-market ratio), equally evidenced past times such models' pitiful functioning during the mesh bubble. This is non surprising because most commonly used factors rely on historical sales or profits measures to approximate companies, land many increase stocks accept really curt history together with piffling or no profits to report. However, equally pointed out lately past times Barry Rehfeld inwards the New York Times, Professor Mohanram of Columbia University has devised a uncomplicated constituent model that rely on viii really convincing factors to score increase stocks. These factors are:

  1. Normalized provide on assets.
  2. Normalized provide on assets based on cash flow.
  3. Cash menses minus mesh income. (i.e. negative of accrual.)
  4. Normalized profits variability.
  5. Normalized sale increase variability.
  6. Normalized R&D expenses.
  7. Normalized working capital missive of the alphabet spending.
  8. Normalized advertising expenses.
By "normalized", I hateful nosotros withdraw to standardize the numbers amongst honour to the manufacture median. To Prof. Mohanram's credit, he claims exclusively that these factors volition generate returns afterward 1 or ii years, non the short-term returns that many traders await constituent models to deliver. The excess annual provide based on buying the grouping of stocks amongst the highest score together with shorting the grouping amongst the lowest score is a skillful 21.4%. Not exclusively does the combined score generate skillful returns, but each private constituent equally good delivers skillful correlation amongst hereafter returns, proving that the functioning is non due to some questionable alchemy of mixing the factors. For example, it makes skillful intuitive feel that extra spending on R&D together with advertising volition boost hereafter profits for increase stocks.

Interestingly, Prof. Mohanram pointed out that most of the out-performance of the high-score stocks tumble out around profits announcements. Hence for those investors who don't similar belongings a long-short portfolio for a sum year, they tin only merchandise during profits season.

One caveat of this enquiry is that it was based on 1979-99 information (at to the lowest degree for the preprint version that I read). As many traders accept works life out, strategies that operate spectacularly inwards the 90's don't necessarily operate inwards the terminal few years. At the really least, the returns are unremarkably greatly diminished. In the future, I promise to perform my ain enquiry to encounter whether this strategy is even together with thus belongings upwards amongst the latest data.

Senin, 29 Juni 2020

An Update On Why Quantitative Funds Are Losing Coin Recently

A even out only came through Dow Jones newswire ("How Black Boxes Became Pandora's Boxes" past times Spencer Jakab) suggesting that recent losses are due to element models gone bad. Given my expressed distaste for such models, that should convey been my begin estimate instead of blaming the "exotic" models!

Jumat, 18 Mei 2007

A Stock Element Based On Selection Volatility Smirk

A reader pointed out an interesting paper that suggests using selection volatility smirk equally a constituent to rank stocks. Volatility smirk is the divergence betwixt the implied volatilities of the OTM position selection as well as the ATM telephone band option. (Of course, in that place are numerous OTM as well as ATM position as well as telephone band options. You tin cite to the master newspaper for a precise definition.) The consider is that informed traders (i.e. those traders who convey a superior mightiness inwards predicting the side past times side turn a profit numbers for the stock) volition predominately purchase OTM puts when they intend the hereafter turn a profit reports volition hold out bad, thence driving upwards the cost of those puts as well as their corresponding implied volatilities relative to the to a greater extent than liquid ATM calls. If nosotros piece of job this volatility smirk equally a constituent to rank stocks, nosotros tin cast a long portfolio consisting of stocks inwards the bottom quintile, as well as a brusk portfolio alongside stocks inwards the summit quintile. If nosotros update this long-short portfolio weekly alongside the latest volatility smirk numbers, it is reported that nosotros volition bask an annualized excess provide of 9.2%.

As a standalone factor, this 9.2% provide may non look terribly exciting, peculiarly since transaction costs convey non been accounted for. However, the beauty of constituent models is that yous tin combine an arbitrary let out of factors, as well as though each constituent may hold out weak, the combined model could hold out highly predictive. H5N1 search of the keyword "factor" on my spider web log volition reveal that I convey talked nearly many dissimilar factors applicable to dissimilar property classes inwards the past. For stocks inwards particular, in that place is a short term factor equally uncomplicated equally the previous 1-day provide that worked wonders. Joel Greenblatt's famous "Little Book that Beats the Market" used two factors to rank stocks (return-on-capital as well as turn a profit yield) as well as generated an APR of 30.8%.

The question, however, is how nosotros should combine all these dissimilar factors. Some constituent model aficionados volition no doubtfulness propose a linear regression fit, alongside hereafter provide equally the subject variable as well as all these factors equally independent variables. However, my sense alongside this method has been unrelentingly poor: I convey witnessed millions of dollars lost past times diverse banks as well as funds using this method. In fact, I intend the only sensible agency to combine them is to but add together them together alongside equal weights. That is, if yous convey 10 factors, but cast 10 long-short portfolios each based on ane factor, as well as combine these portfolios alongside equal capital. As Daniel Kahneman said, "Formulas that assign equal weights to all the predictors are ofttimes superior, because they are non affected past times accidents of sampling".


Kamis, 10 Mei 2007

Variance Guide Chances Premium For Furnish Forecasting

Folklore has it that VIX is a reasonable leading indicator of risk. Presumably that way if VIX is high, in addition to therefore in that place is a expert adventure that the futurity provide of the SP500 volition last negative. While I convey found simply about bear witness that this is truthful when VIX is peculiarly elevated, enjoin inwards a higher house 30, I don't know if anyone has established a negative correlation betwixt VIX in addition to futurity returns. (Contemporaneous VIX in addition to SP500 levels create convey a really prissy linear human relationship amongst negative slope.)

Interestingly, the province of affairs is much clearer if nosotros examine the Variance Risk Premium (VRP), which is defined every bit the difference betwixt a model-free implied volatility (of which VIX is the nearly famous example) in addition to the historical volatility over a recent period. The human relationship betwixt VRP in addition to futurity returns is examined inwards a newspaper by Chevallier in addition to Sevi in the context of OVX, which is the CBOE Crude Oil Volatility Index. They convey found that in that place is a statistically pregnant negative linear human relationship betwixt VRP in addition to futurity 1-month unsmooth crude oil futures (CL) returns. The historical volatility is computed over 5-minute returns of the nearly recent trading day. (Why v minutes? Apparently this is long plenty to avoid the artifactual volatility induced yesteryear bid-ask bounce, in addition to curt plenty to genuinely sample intraday volatility.)  If you lot believe inwards the prescience of options traders, it should non surprise you lot that the regression coefficient is negative (i.e. a high VRP predicts a lower futurity return).

I convey tested a elementary trading strategy based on this linear relationship. Instead of using monthly returns, I role VRP to predict daily returns of CL. It is really similar to a mean-reverting Bollinger ring strategy, except that hither the "Bollinger bands" are constructed out of moving get-go in addition to tertiary quartiles of VRP amongst a 90-day lookback. Given that VRP is far from usually distributed, I idea it is to a greater extent than sensible to role quartiles rather than measure deviations to define the Bollinger bands. So nosotros purchase a front end contract of CL in addition to concur for simply 1 twenty-four hours if VRP is below its moving get-go quartile, in addition to curt if VRP is inwards a higher house its moving tertiary quartile. It gives a decent average annual provide of 17%, but functioning was misfortunate inwards 2013.

Naturally, i tin crusade this elementary trading strategy on the E-mini SP500 futurity ES also. This time, VRP is VIX minus the historical volatility of ES.  Contrary to folklore, I respect that if nosotros regress the futurity 1 twenty-four hours ES provide against VRP, the regression coefficient is positive. This way that an growth of VIX relative to historical volatility really predicts an growth inwards ES! (Does this hateful investors are overpaying for set options on SPX for portfolio protection?) Indeed, the contrary trading rules from the inwards a higher house plough over positive returns: nosotros should purchase ES if VRP is inwards a higher house its moving tertiary quartile, in addition to curt ES if VRP is below its moving get-go quartile. The annualized provide is 6%, but functioning inwards 2013 was too poor.

As the authors of the newspaper noted, whether or non VRP is a potent plenty stand-alone predictor of returns, it is in all likelihood useful every bit an additional component inwards a multi-factor model for CL in addition to ES. If whatever reader know of other volatility index similar VIX in addition to OVX, delight create percentage amongst us inwards the comments section!

===

My online Backtesting Workshop volition last offered on Feb 18-19. Please visit epchan.com/my-workshops for registration details. Furthermore, I volition last pedagogy my Mean Reversion, Momentum, in addition to Millisecond Frequency Trading workshops inwards London on March 17-21, in addition to inwards Hong Kong on June 17-20.

Rabu, 09 Mei 2007

Fundamental Factors Revisited, Amongst A Applied Scientific Discipline Update

Contrary to my tradition of alerting readers to novel together with fancypants factors for predicting stock returns (while non necessarily endorsing whatever of them),  I study that Lyle together with Wang convey lately published novel interrogation demonstrating the ability of 2 real familiar factors: book-to-market ratio (BM) together with return-on-equity (ROE).

The model is simple: at the terminate of each calendar quarter, compute the log of BM together with ROE for every stock based on the around recent profits announcement, together with regress the next-quarter render against these 2 factors. One subtlety of this regression is that the ingredient loadings (log BM together with ROE) together with the futurity returns for stocks inside an manufacture grouping are pooled together.  This makes for a cross-sectional ingredient model, since the ingredient loadings (log BM together with ROE) vary past times stock but the ingredient returns (the regression coefficients) are the same for all stocks inside an manufacture group. (A clear elucidation of cross-sectional vs time-series ingredient models tin terminate hold out institute inward Section 17.5 of Ruppert.) If nosotros long stocks inside the overstep decile of expected returns together with curt the bottom decile together with concur for a quarter, the expected annualized average returns of this model is an eye-popping 26% or so.

I convey tried to replicate these results, but unfortunately I couldn't. (My programme generated a measly, though positive, APR.) The information requirement together with the programme are both quite demanding. I am unable to obtain the sixty quarters of telephone substitution information that the authors recommended - I simply convey 40. I used the 65 manufacture groups defined past times the GIC manufacture classifications, spell the authors used the 48 Fama-French manufacture groups. Finally, I am unsure how to bargain alongside stocks which convey negative mass values or earnings, together with thus I omit those quarterly data. If whatever of our readers are able to replicate these results, delight exercise permit us know.

The authors together with I used Compustat database for the telephone substitution data. If you lot exercise non convey subscription to this database, you lot tin terminate regard a new, free, website called Thinknum.com. This website makes available all information extracted from companies' SEC filings starting inward 2009 (2011 for small-scale caps). There is also a great integration alongside R described here.

*** Update ***

I forgot to betoken out i essential divergence betwixt the method inward the cited newspaper together with my ain effort: the newspaper used the entire stock universe except for stocks cheaper than $1, spell I did my interrogation alone on SP500 stocks (Hat tip to Prof. Lyle who clarified this). This turns out to hold out of major importance:  a to-be-published newspaper past times our reader I. Kaplan reached the determination that "Linear models based on value factors exercise non predict futurity returns for the S&P 500 universe for the past times 15 years (from 1998 to 2013)."


===

Speaking of novel trading technology scientific discipline platforms that supply historical information for backtesting (other than Thinknum.com together with the previously mentioned Quantopian.com), hither is roughly other interesting one: QuantGo.com. It provides institutional intraday historical information through its information partners from 1 infinitesimal bars to sum depth of mass inward your ain soul cloud running on Amazon EC2 concern human relationship for a depression monthly rate. They give unlimited access to years of historical information for a monthly information access fee, for examples States equities Trades together with Quotes (TAQ) for an unlimited issue of years are $250 per calendar month of concern human relationship rental, OPRA TAQ $250 permonth together with tagged tidings is $200. Subscribers command together with contend their ain calculator instances, together with thus tin terminate install together with role whatever software they desire on them to backtest or merchandise using the data. The alone hitch is that you lot are non allowed to download the vendor information to your ain computer, it has to remain inward the soul cloud.

===

Follow @chanep to have my occasional tweets on interesting quant trading manufacture tidings together with articles.

===

My online Mean Reversion Strategies Workshop volition hold out offered on Apr 1-3. Please view epchan.com/my-workshops for registration details. Furthermore, I volition hold out instruction my Mean Reversion, Momentum, together with Millisecond Frequency Trading workshops inward London on March 17-21, together with inward Hong Kong on June 17-20.

Selasa, 08 Mei 2007

Update On The Fundamentals Factors: Their Upshot On Pocket-Size Cap Stocks

In my terminal post, I reported that the key factors used past times Lyle in addition to Wang look to generate no returns on SP500 large cap stocks. These key factors are the increase constituent return-on-equity (ROE), in addition to the value constituent book-to-market ratio (BM).

I convey since studied the outcome of these factors on SP600 pocket-size cap stocks since 2004, using a survivorship-bias-free database combining information from both Compustat in addition to CRSP. This time, the factors gain create an annualized average render of 4.7% in addition to a Sharpe ratio of 0.8. Though these numbers are nowhere close the 26% render that Lyle in addition to Wang found, they are withal statistically significant. I convey plotted the equity crimp below.

 I reported that the key factors used past times Lyle in addition to Wang look to generate no returns Update on the fundamentals factors: their outcome on pocket-size cap stocks
Equity crimp of long-short small-cap portfolio based on regression on ROE in addition to BM factors (2004-2013)
One may wonder whether ROE or BM is the to a greater extent than of import factor. So I run a simpler model which uses ane constituent at a fourth dimension to rank stocks every day. We purchase stocks inward spill out decile of ROE, in addition to curt the ones inward the bottom decile. Ditto for BM. I constitute an annualized average render of 5% alongside a Sharpe ratio of 0.8 using ROE only, in addition to only 0.8% alongside a Sharpe ratio of 0.09 using BM only. The value constituent BM is almost completely useless! Indeed, if nosotros were to starting fourth dimension form on ROE, selection the spill out in addition to bottom deciles, in addition to hence form on BM, in addition to selection the spill out in addition to bottom halves, the resulting average render is almost the same equally sorting on ROE alone. I plotted the equity crimp for sorting on ROE below.

Equity crimp of long-short small-cap portfolio based on spill out in addition to bottom deciles of ROE (2004-2013)

Notice the sudden drawdown from 2008-05-30 to 2008-11-04, in addition to the almost perfect recovery since then. This mirrors the demeanour of the equity marketplace itself, which raises the interrogation of why nosotros bother to build a long-short portfolio at all equally it provides no hedge against the downturn. It is likewise interesting to banking concern complaint that this constituent does non demo "momentum crash" equally explained inward a previous article: it does non endure at all during the marketplace recovery. This agency nosotros should non automatically recollect of a key increase constituent equally similar to cost momentum.

My determination was partly corroborated past times I. Kaplan who has written a preprint on a similar topic. He constitute that a long-short portfolio created using the ratio EBITA/Enterprise Value on large caps generates a Sharpe ratio of close 0.6 but alongside real picayune drawdown dissimilar the ROE constituent that I studied higher upwardly equally applied to pocket-size caps.

As Mr. Kaplan noted, these results are inward roughly contradiction non only alongside Lyle in addition to Wang's paper, but likewise alongside the widely circulated paper past times Cliff Asness et al. These authors constitute the the BM constituent industrial plant inward practically every property class. Of course, the timeframe of their inquiry is much longer than my focus above. Furthermore, they convey excluded fiscal in addition to penny stocks, though I did non uncovering such restrictions to convey non bad affect inward my study of large cap portfolios. In house of a key increase factor, these authors only used cost momentum over an 11-month stream (skipping the most recent month), in addition to constitute that this is likewise predictive of futurity quarterly returns.

Finally, nosotros should banking concern complaint that the ROE in addition to BM factors hither are quite similar to the Return-on-Capital in addition to Earnings Yield factors used past times Joel Greenblatt inward his famous "Little Book That Still Beats The Market". One wonders if those factors endure a similar drawdown during the fiscal crisis.

===

My online Momentum Workshop volition live on offered on May 5-7. Please watch epchan.com/my-workshops for registration details. Furthermore, I volition live on educational activity my Mean Reversion, Momentum, in addition to Millisecond Frequency Trading workshops inward Hong Kong on June 17-20.


Senin, 07 Mei 2007

Short Involvement Equally A Factor

Readers of zerohedge.com volition no dubiety hold upward impressed yesteryear this chart and the accompanying article:

Cumulative returns of most shorted stocks relative to SPX Short Interest every bit a Factor
Cumulative Returns of Most Shorted Stocks inward 2013

Indeed, brusk involvement (expressed every bit the publish of shares shorted divided yesteryear the full publish of shares outstanding) has long been idea to hold upward a useful factor. To me, the counter-intuitive wisdom is that the to a greater extent than a stock is shorted, the ameliorate is its performance. You mightiness explicate that yesteryear maxim this is a lawsuit of the "short squeeze", when at that spot is confine inward cost perhaps due to intelligence in addition to stock lenders are eager to sell the stock they own. If you lot accept borrowed this stock to short, your borrowed stock may hold upward recalled in addition to you lot volition hold upward forced to purchase comprehend at this most inopportune time. But this is an unsatisfactory explanation, every bit this volition lawsuit solely inward a brusk term (upward) momentum inward price, non the sustained out-performance of the most shorted stocks. This long-term out-performance seems to propose that brusk sellers are less informed than the average trader, which is odd.

Whatever the explanation, I am intrigued to abide by out if brusk involvement actually is a skilful component to contain into a comprehensive component model over the long term.

The result? Not peculiarly impressive. It turns out that 2013 was ane of the best years for this component (hence the impressive nautical chart above). For that year, a daily-rebalanced long-short portfolio (long fifty most shorted stocks in addition to brusk fifty to the lowest degree shorted stocks inward the SPX) returned 6.9%, amongst a Sharpe ratio of 2 in addition to a Calmar ratio of 2.9. However, if nosotros extend our backtest to 2007, the APR is solely 2.8%, amongst a Sharpe ratio of 0.5 in addition to a Calmar ratio of 0.3. This backtest was done using survivorship-bias-free information from CRSP, amongst brusk involvement information provided yesteryear Compustat.

Here is the cumulative returns nautical chart from 2007-2013:

Cumulative Returns of LS Portfolio based on Short Interest: 2007-2013


Interesting, trying this on the SP600 small-cap universe yielded negative returns, perhaps important that short-sellers of pocket-size caps produce accept superior information.

I promise, this volition hold upward the final fourth dimension I verbalize almost factors inward a while!

===
Tech Update:

I was shocked to larn that Matlab straightaway offers licenses for simply $149 - the so-called Matlab Home  (h/t: Ken H.) In addition, its Trading Toolbox straightaway offers API connectedness to Interactive Brokers, inward add-on to a few other brokerages. I am familiar amongst both Matlab in addition to R, in addition to spell I am impressed yesteryear the large publish of free, sophisticated statistical packages inward R, I withal stand upward yesteryear Matlab every bit the most productive platform for developing our ain strategies. The Matlab evolution (debugging) surroundings is simply that much to a greater extent than polished in addition to easy-to-use. The departure is bigger than Microsoft Word vs. Google Docs.
Influenza A virus subtype H5N1 reader Ravi B. told me that at that spot is a website called www.seasonalgo.com if you lot desire to crusade out dissimilar seasonal futures strategies.
Finally, a startup at inovancetech.com offers machine learning algorithms to aid you lot abide by the best combination of technical indicators for trading FX.
===
Workshops Update:

I am straightaway offering the Millisecond Frequency Trading (MFT) Workshop every bit an online course of report on June 26- 27. Previously, I accept solely offered it alive inward London in addition to to a few institutional investors. It has 2 principal parts:

Part 1: introducing techniques for traders who desire to avoid HFT predators.

Part 2: how to backtest a strategy that requires tick information amongst millisecond resolution using Matlab.

The illustration strategy used is based on lodge flow. For to a greater extent than details, delight catch epchan.com/my-workshops.

Additionally, I volition hold upward didactics the Mean Reversion in addition to Momentum (but non MFT) workshops inward Hong Kong on June 17-20.