Legal Front Running

evanscjeUncategorized0 Comments

Yes, that’s the term for putting in an order ahead of a client order. It is illegal but the idea is exciting. What if we could figure out a way to do it – legally?
Who are the victims?
We want to victimize slow moving competitors – if it helps us. The idea is to buy the market after a bout of trauma before everyone else buys it back. We are assuming that such trauma forces many participants out of the market through stops or panicky committee decisions. If we want to make money from the long side, then we are looking to take advantage of slower moving investors. We will include in our analysis the 2008 experience where there was a considerable delay before the rise in fear rewarded buyers.
The Trauma Measure
We’ll measure fear using a version of my variable – eFear. I wrote about the result of simply buying when eFear was above a 20-day average but here we need to define real trauma. Let’s say that the if eFear rises above its monthly average plus 2.5 standard deviations, then market participants will be very uncomfortable. This occurred 69 times over the last 15 years. To get a clearer idea of these events, I show the two series below:

If the trauma hurdle is too low, then we’ll get too many inconsequential events. If they are too rare, we’ll never get a chance to buy the market. Five per year seems like a reasonable number.
Day Counts
The day of the crossover we shall call day zero. After that, the count will go up by one each day unless eFear crosses above that level again. The idea is that a low number should be good. The closer we are to the event, fewer shorts have been covered and most cash is still unspent. The level of eFear post trauma is not a consideration. It frequently corrects very quickly and our goal here is to see how we do if we ignore the actual level of fear.

The eFear level is not entirely intuitive because it is compared to volatility. Fear must rise faster than volatility. This chart gives you a good idea of how we will spend most of our time relatively far away from such events.
A Trading Test
Now we get to play one of my trading games: Paratrade will get to hold the market for any day count <20 and the client will hold it when the count is >=20. Since 2002 Paratrade held the market for 1179 days and the client held it for 2587 days. Paratrade made 750 points and the client made 60. But it’s worse than that – for the client. The worst drawdown was 200 points for Paratrade vs 700 for the client!
The problem here may be that you can’t wait for long periods to buy the market. Customers and managers demand that you “participate” more. You need more trauma events to occur so we need a tighter definition. Let’s set the bar to the average +2 standard deviations. Now we can be even pickier. We have more buying opportunities so we don’t need to hold a position for so long. We’ll set the count maximum to ten. Now Paratrade holds the market for 1205 days and the client holds it for 2561. Over almost 15 years, Paratrade made 554 points and the client made 257. In 2008, the Paratrade drawdown was half of the client’s.
Part of the reason for its success is that when you get a bad signal and the market is trending down, the longest Paratrade will hold it for is ten days and then it requires a new crossover. In 2002, 2008 and 2011 drawdowns were reduced and the system allowed for easy reentry.

This is a very damning statement about trend following. If you wait for the market to recover, you end up missing most or all of the move. If your breakout or trend following solution requires lagging variables to confirm, you’ll be buying just before you should be exiting.
The worst part is that the further away you get from any bout of trauma, the closer you are to another one. This means your trend following system will have to hold up through a correction just after it got long. Smooth extended rallies in equities occur rarely and it turns out that if you miss them, it’s no big deal.

Leave a Reply

Your email address will not be published. Required fields are marked *