Several of my algorithm ideas that have survived out-of-sample test have been pairs strategies focusing on exchange-traded products. I might tell you a bunch of stuff you already know here, so bear with me. Many ETPs are convoluted, are "not meant to be held more than a day", or are apparently marketed only to retail investors. Examples are:
1) Inverse funds such as SH. If you short both SPY and SH on Quantopian, you will see a square-wave pattern with a quarterly period -- this corresponds to the dividend cycle of SPY and SH, one paying out, one paying in, when both are short-sold. In addition to the square-wave pattern, shorting both gives a mean return of about 1% per year. This is due to the fee difference between the funds.
2) Futures-backed commodity ETNs like WEAT. I think WEAT is one of the worst funds on the exchange, with a whopping expense ratio of 3.34%. WEAT can be hedged by going long on consumer staples with a fund like XLP.
3) Volatility ETNs like VXX. Volatility ETNs would be great to short because they often come with explicit inverses. VXX can be hedged by simultaneously going short on VXX and XIV.
4) Leveraged funds. Obviously leveraged funds are not permitted in the competition (I guess I should have taken this as a clue).
5) Other funds with high fees. One thing common to 1-4 above is that they charge abnormally high fees. This points out a fifth strategy, namely just a long-short strategy based on ETF management expense ratios.
Unfortunately I now suspect that most if not all of the funds listed above would not be desirable to short, because either huge collateral would be needed, a high fee would be charged, or the broker may not actually be able to locate the shares to borrow. I found this out from a few threads on the forums, where people were talking about shorting leveraged ETNs in a similar fashion:
I don't know whether this totally sinks all of (1-5) above, or if it just reduces their profitability. What do you think?