Back to Community
I really wish they allowed the use of leveraged ETFs (7.18 sharpe, 13.65 sortino, -0.02 beta, 0.01 volatility)

Image of my results

Leverage ratio should be determined using net leverage...

Consider the following example:

3x leverage in USO = 3x net leverage

3x leverage in UWTI and
3x leverage in DWTI = 0x net leverage...

23 responses

I'm sure many have similar feelings about the 3x levered ETF's. 2 months of back test isn't very robust IMO though. Those etfs have traded much longer, why not use a longer backtest?

Nice. I have a strategy that tries to arb the tracking errors from volatility in leveraged ETF pairs, it has similarly stable in returns and is entirely market neutral. There is a lot of missed opportunity in the contest for what seems to be an easy fix to me.

If your strategies involve shorting those 3x funds, be aware that the annual short fee rates are in the 5-10% range, and you cannot leverage the positions.


That one does something similar. The key is to predict volatility. This can be done by looking at the probability of informed trading. Also done through looking at correlated securities or underlying securities.


IB account holders within certain tiers also receive a rebate for the interest on the cash position that results from the short position. In the securities I am interested the fees are negligible.

Well aware, IB has protocol to generate a CSV of current short availability as well as borrow fees. To receive the rebate though, I believe your account must have a substantial borrowed position though I could be wrong, this is the case with most brokers.

I believe the spread is between 0.25% and 2% depending on the size of the position. The strategy is scalable enough to get the 0.25% spread

Yes, but cash interest rebates are negligible at the moment, vs short fees. Anyway, I'm sure you guys know all this. has some data.

Wow I wish I would have found that paper when I started building that strategy months ago. It's somewhat reaffirming in that I came to many similar conclusions through my own analysis, mainly that the majority of the time this is not a profitable position. I think where the 'secret sauce' is, is in identifying which pair to be arbitraging at any given time.
As the paper suggests, there is a small portion of these LETF pairs which have a positive expected return at a 90% CI, the issue is that which LETFs are profitable will change over time. So if you can build a filter to identify the profitable pairs vs. not, you have a profitable strategy.
Thanks for the link, definitely bookmarking that one.

Yes, if one had a consistent method of predicting which pairs were going to experience that effect before they do, that would be very handy. I have identified some correlates, but haven't had success yet in making any predictive models/filters.

Anthony Ng added a link to the Ideas thread which is relevant here:

A good strategy should work regardless of leverage

It does work without leverage. Allowing the use of leveraged ETFs simply opens up quantopian competitors to more possibilities for conducting statistical arbitrage.

So I'm going to push back on the "A good strategy should work regardless of leverage" assertion that was made with nothing to support it. I've got plenty of hedged leveraged strategies with sub-2% max drawdowns over many years, the corresponding very low volatility, and a beta near 0 that yield unimpressive returns without leverage but very decent returns with it, considering the beta and volatility. Its a bit presumptuous to simply dismiss this entire line of algorithms as bad strategy simply because they require leverage to work, no?
Back to the original topic, I fully agree with the OP. I'd love to see leveraged ETFs precisely because of all the bad things they having going for them and not at all for their leverage. I've spoken offline with Quantopian extensively on this and I think they have some good points as to why they don't have them at least on the short side and it has little to do with leverage. Many of these are very hard to borrow, which means they either aren't really available to short when your backtester assumes they are and/or that they'll get recalled randomly over time if you hold them more than three days. The historical data on this probably doesn't even exist, but certainly isn't in the standard databases Q has access to. In addition, as mentioned they often have a very high borrow rate (we're talking 50% plus I've seen on some!), and that changes daily. Q doesn't have access to the historical daily borrow rates, again an arcane non-standard data set, so they can't simulate that in backtesting. Both of these are kind of showstoppers for the contest, but nothing's stopping us from setting up live trading accounts with a few thousand dollars of real money and running the algorithms from there. I guess my only request might be to ask Q to allow live trading accounts with more than two months of history to compete in the contest since the paper trading algorithms are going 100% off their trading record by that point and the real money account would if anything be more realistic than a paper trading one. I'm personally still working an IB paper trade account until I can develop the confidence in both IB and Quantopian to put real money on the line, but knowing my live account will compete in the contest might be enough to get me to jump.


I very much agree with you; given the problems you mentioned, the submission of live trading algorithms is a great solution.

I am also sticking to IB paper trading for the same reasons. I've had enough success doing that to where I believe my strategies should work even with borrowing costs etc..

I've designed my algo so that it only conducts the arb for short periods when I expect high volatility. There's some trickiness to keeping trading costs from getting too ridiculous. That involves predicting when HFTs reach across the spread...

Anyways it means the borrowing costs are at least insignificant enough that the strategy is still profitable and that the time restraint is irrelevant.

I have also used my IB real money account to execute a simple version of my statistical arb strategy manually. I've had success doing that too, but it's simply way too time consuming, difficult to execute properly, and ultimately unsustainable.

I have faith that Q will figure all this out, but I'm also anxious to see it happen sooner rather than later. I think your suggestion is a very plausible solution.

@ Kevin,

It comes down to by what factor the returns outperform the drawdowns (if they do at all). That, combined with your appetite to take losses determines how much leverage you can use. I'm not saying you shouldn't leverage. I'm only saying leverage is a double edged sword. When you increase your leverage, you are essentially increasing all the risk metrics - VaR, drawdown, violatility of returns, beta (if it is non-zero) etc. If these metrics are healthy even after increasing leverage, no problem!


Trading leveraged ETFs is not synonymous with taking on leverage. We're talking about conducting arbitrage. This means being hedged.

I've come to the conclusion that this may be one of those times that those of us who get it are better off if those who don't, don't.

I came to the same conclusion as you about trying to be in the arb just a few days to get around the short issues. The problem I run into here is my only other big gripe against Quantopian, their required commission and slippage model for the contest that is simply not based in any kind of reality.They've promised to replace that with something that more closely models real commission at least, so hopefully after the current software glitch is resolved they'll move on that. The good news is it probably only requires them to change a couple lines of code.

Urg I agree with you Kevin, but I feel compelled to give Bharath an example.

Suppose you invest 100% of your account in SPY, a 1x ETF. You are not leveraged, in any sense of the word, but you are fully exposed to every swing of the market, and have quite a lot of risk.

Suppose instead you invest 100% in SPY and borrow another 100% to invest in SH (inverse 1x S&P 500 ETF). You are now 200% leveraged, but your exposure to the market is, for today, 0. You are perfectly hedged, and you have very little risk, even though your leverage has doubled. You might need to rebalance these positions periodically to make sure that your notional beta is still 0, but you're much safer than before.

Suppose now you invest 50% of your account in SSO (2x S&P 500 ETF) and 50% of your account in SDS (inverse 2x S&P 500 ETF). Adding it all up, you are still 200% leveraged, and your exposure to the market is, for today, 0. Nothing has changed, except some details related to the convexity of your position, you are long some gamma and you must rebalance more frequently if you want to stay hedged.

This is also true if you short -50% of SSO and -50% of SDS, which are the sort of positions I believe everyone on this thread is talking about. Leveraged ETFs haven't added any obvious risks, except in this case, now you're short gamma, which may or may not lead to an eventual drawdown/blowup. Who knows.

But one thing that IS true, is that leverage hasn't, in these cases, magnified anything.

Also Kevin/Taylor: it's worth trying these positions out on real money, to get a feel for when the shorts are hard to find. I am discovering, and am therefore a little worried, that the short locate problems happen at precisely the time when you need them most and the market is moving fast. It seems to be a "wrong-way risk", which might make it hard to tactically get into these positions when the market starts to move.

I think you're definitely right. I'll be interested to see my results when I finally pull the trigger but I have the same feeling as you. On the plus side, if it was easy everyone would be doing it!

Taoylor, I agree. If you are hedging, it doesn't matter whether you are trading levered instruments or unlevered ones. As long as your historical max drawdown doesn't hit the roof, it means your hedging is working.

quantopian isn't calculating the shapre correctly . also going back to 2013 gives a much lower sharpe