55% CAGR with 40% drawdown is really impressive for a short only strategy. Do you mind sharing what universe, start/end period, and rebalancing frequency?
I had been running a multifactor short strategy through p123 in a linked interactive brokers account, but I ended up winding it down towards the end of 2024 as there were significant headwinds.
As you and Yuval pointed out, it can be really difficult when factor performance inverts and both your long and short legs bleed money. Furthermore, since many multifactor systems typically include factors like low beta and low volatility, then shorting strategies based on the same ranking system will favor high beta and high volatility stocks. So when the markets go up with factors inverted, you can lose a little bit of money on your low vol longs and a lot of money on your high vol shorts, which is not very palatable for what is supposed to be your hedge.
This played out a couple times in 2024, like in March when the system shorted SMR when nuclear stocks were going, well, nuclear. And again in November when the system wanted to short QUBT, ATOM, and MSTR when quantum computing and semiconductor stocks were rallying like crazy as was bitcoin. Fortunately by that point, I had already been overriding most of the trades before capitulating altogether.
Other obstacles I encountered:
- High actual borrow costs: As I was using a broad US trading universe, the shorting system seemed very good at selecting stocks with higher borrow rates. My actual borrow costs hovered in the 10-15% range and that was even after manually filtering trades with very high (>50% or sometimes even 100+% annualized borrowing costs). I didn't have enough data to assess the relationship between actual borrow costs and expected value of the trade.
- No historical borrow costs: the lack of borrow costs in p123 is a significant impediment to accurately modeling sim costs. How can you accurately compare two candidate short sims yielding X% and Y% if you don't account for borrow costs?
- Performance tracking: There is an outstanding bug with managed short strategy performance that makes it more difficult to accurately track performance of live trading to sim.
- Separate construction of long/short portfolios: makes it hard to make sector or industry neutral bets.
- Monthly risk measurements: most places in the p123 UI show risk measures like the Sharpe ratio computed with monthly samples (I believe). You can see daily and weekly measures if you drill down on the Statistics -> Risk Measurements tab, but the default monthly statistics can obscure some of the turbulence that is really ocurring at the day-to-day level.
There's an adage that the real litmus test of of whether you're taking on too much risk in your portfolio is if you have trouble sleeping at night. Ultimately, that's what was happening with this short strategy -- which was supposed to be my hedge, too. I'd have to continually check changing borrow costs and meme stock lists to make sure I wasn't going to get run over.
I'm currently working on some refinements to some long strategies, but I plan to retool the short leg in the future. I have some promising preliminary sim results restricting the shorting to a more liquid universe like the S&P500, Russell 1000, or even S&P 1500 which tend to be more widely easy to borrow and less prone to retail investor mania. But I'm still not sure if I'd be ready to trade this through p123 with some of the performance tracking and portfolio construction limitations that currently exist -- I may have to export my ranks to some other system.