NuclearPhynance Archive

of course is possible, depends on size i think.



@EspressoLover answer is more suited to big PA'a where having a day job when it's not running your company/companies not makes much sense any more.



but i can be wrong on that as for someone else different things can work.

My view has always been that over the long run, beta is good enough for PA. Except when I am unemployed - trading the PA beats watching daytime TV.

> I am not sure what you mean by relatively simple implementation



Obviously I'm not giving this advice to my dentist. But I think if you can 1) program medium-sized scripts, 2) read and understand quant finance papers, and 3) understand the markets with some level of professional intuition, it's not that hard. So, pretty much every NP poster.



If your aim is just to replicate well-known premia, there's a lot of things working in favor of simple implementation:



* You don't have to backtest all. That's already been done for you by the researchers. You just have to construct present-day portfolios.

* That also means you don't have to worry about historical data, and all the associated cost and issues. You only need contemporaneous data, or very recent historical data.

* If a factor's so well known, chances are that it's input data is widely available for cheap or free. Book value, market cap, market beta, earnings calendars, are widely disseminated. Biggest pain in the ass is maybe writing a web-scraper.

* Reconciliation is pretty easy, because the major factors usually have published returns which are frequently updated somewhere (Ken French, AQR, etc.). Just make sure your recent live returns are near in line with some source of truth.

* Turnover on most of the classical premia tends to be pretty low. Most only rebalance monthly, and even then there's usually not much turnover. You don't need a spiffy automated system that continuously trades. Just dump trades at the end of the month, eyeball for correctness and send a batch of orders.

* T-Costs in major developed markets are basically nothing at this horizon. Unless you're very fortunate, your PA doesn't have to worry about market impact.

* In most developed markets, short selling is easy and pretty cheap.

* A typical Decile[1 Minus 10] factor on the Russell 3000, takes positions in 600 names. Obviously that's not feasible. But taking an unbiased subset of 10-15 long and short symbols is going to approximate the portfolio with pretty low variance. Unless you're a major portfolio, 10-15 names is more than enough liquidity.

* Historical performance is well-known. Just apply a sensible shrinkage estimator to expected long-run forward returns. All of the major premia are pretty much orthogonal to each other and the market. So portfolio allocation is dead-simple.

* The exact details of most factors are pretty meaningless. They're pretty much just Schelling points, decided by the arbitrary decisions of the first academic to publish on the topic. These aren't ultra-fragile strategies that require precise execution. So even if your implementation gets a few details wrong by accident or necessity, it's unlikely to have a significant impact.



The biggest challenge is just being disciplined. When a factor under-performs for five years, that's sixty rebalances of bad vibes. There's a strong psychological impulse to abandon it (which is a large part of why these things tend to keep working). There's also the temptation to tinker. "Oh, well it seems like [X] doesn't work when [Y], so I'm going to add this modifier or filter". But once you do start doing that, you're now trying to generate alpha. All those simplifying conditions go out the window, and unless it's your full time job, it's probably not going to work out well for you.

Eq smart beta / alt premia strategies are quite readily available nowadays in even ETF form, albeit long only. diy l/s has some potentially unpleasant / time consuming aspects though (corporate actions, so annoying). I have for several years traded a watered down versions of my futures strategies using ETFs (sadly can't trade futures in PA). 30 or so ETFs, borrow costs can sometimes suck but hey, half of a 2.5 sharpe strategy is still decent for PA

EL, I think your proposal sounds a bit like you are trying to compute a Fourier series of a function, but you don't know the coefficients at all and you only know sines and cosines very approximately, yet you somehow hope that you are going to get a good approximation to the function. Get real!

Ok.



Now let's assume I do have strategy with a sharp 1.5 plus....(well, very hand waving with calculation\estimation of transaction costs...but shouldn't be critical here)



Rebalanced monthly. Based on a set of 26 ETFs. Basicaly a diversified momentum thingy.



My problem is: I am reluctant to invest a significant percentag of my money only in ETFs. Everything that growths that massiveley in size like the invested volume in ETFs scares me.



What I mean, my invetsment strategy appears to be quite "market neutral" if you take a major equity index as "market".



I am worried that the diversification/hedging effect of the strategy (short ETFs, ETFs on asset classes) breaks down if markets crash and the ETF industry (not the equity ETFs...) with it?



Any thoughts on that? Or is it possible to detect things like a priori that an stay on the sidelines?

@maggette,



The short answer is that nobody knows because it hasn't happened yet. Some people have some theories (e.g. benefits of sticking with physical etfs vs synthetic etfs etc) but the reality is that that won't help all that much if liquidity dries up like it did in 2008. Correlations also all break down when everybody rushes for the exit.



So you are left with the strategy for any crash - keep your leverage low going into the crash, and high coming out of it.



A bigger worry would be the cost of shorting your etfs. That's what kills market neutral etf strategies most of the time.

I haven't looked into it in detail, but has anybody calculated the effects of US taxes on some person strategy like what EspressoLover suggests?



1) Assuming the bulk of your portfolio is not in tax deferred accounts, the turnover is subject to short term capital gains, or even long term capital gains can knock 10-20% off your profits each year. If you are only squeezing out 12%, the taxes could drop it to 9%. then maybe it is not a win.



2) If you are doing it in Tax-deferred, no shorting. Also, you have to wait for funds to settle before trying again. That keeps you out of the market for 10% of the month. Is that worth it?



Are we benchmarking against the 8% gains of the S&P and hold it for 30 years and paying taxes only on dividends?

@darkmatters



Good questions. This paper has a deeper dive into the tax efficiency of the major equity anomalies.



The major "fundamental anomalies" have pretty low turnover. On the order of 17% per year, which doesn't put it that far away from the S&P 500. On the short leg, you pay ordinary rate, even if you hold for over 12 months, so they're not quite as tax efficient as S&P500. OTOH, most of these anomalies accrue almost all their gains on the long leg (which makes sense, as the market rises over time).



The "trading anomalies" also aren't as tax-bad as they seem at first glance. The turnover's high. But most tend to generate substantial tax-loss realizations. E.g. MOM is repeatedly selling losers and buying winners. Not only does this tend to defer tax realization and concentrate gains in 20% long-terms cap-gains, but it generates very valuable short-term tax loss carry-forwards which can be used in other parts of your portfolio.



Futures and forex are taxed at a blended rate of 28% regardless of holding period. Implementing an anomaly in commodities, rates or FX space (e.g. TSMOM or carry) means that even a tax-naive strategy is still relatively efficient. Futures, even shorts, can be utilized in an IRA. So, for example if you want HML in an IRA, you can buy the long equity leg, then beta-neutralize using ES. The research indicates that the single leg still generates sizable returns for most anomalies. The low-turnover anomalies would only be minimally affected by the 3-day waiting period.



Finally, if you're not liquidity constrained, and using random subsets as approximating portfolios, there's potential tax efficiencies with modest bias-free modifications. Tax losses can be realized and gains deferred by re-sampling on the losing leg. E.g. if the market is up for the year, pick a new subset on the short leg, and keep the subset for the long leg.

I'm trying replicate AQR factors with ETFs, but can't make sense of the data.



For example, I expect HML can be replicated by a value ETF like VLUE, or QMJ by QUAL. But the returns look totally different. Guessing this is because the factors are beta hedged, whereas the ETFs are intrinsically long.



What's an easy way to map the factor exposures to ETFs?