I finally got around to programming an easy Python down loader for CME data from Quandl. I promised this in my last book, but haven’t got that many requests. On the Ultimate Algorithmic Trading System…. page I have provided some data going back to early 2008 on several markers. This is a work in progress. I also wrote a back adjuster that rolls when the Open Interest + Volume is greater in the next contract and adjusts the data retroactively (Panama process) by the discount on the roll date. As I state on the web page this data is not all that great even though it looks like it comes directly from CME. I do fill in gaps and try to fix glaring errors so take it for what it is. Its good data for preliminary testing, but to finalize an algorithm or to trade by I would definitely go the “paid” route. Although the later data looks pretty good.
Before downloading the data I would sign up for Quandl and get an API key. You will be amazed at the amount of data they make available – free and paid subscriptions.
Look at this equity curve:
As long as you are in a bull market buying dips can be very consistent and profitable. But you want to use some type of entry signal and trade management other than just buying a dip and selling a rally. Here is the anatomy of a mean reversion trading algorithm that might introduce some code that you aren’t familiar. Scroll through the code and I will summarize below.
I am using a very short term RSI indicator, a la Connors, to initiate long trades. Basically when the 2 period RSI dips below 30 and the close is above the 200-day moving average I will buy only if I am not buying “a falling knife.” In February several Mean Reversion algos kept buying as the market fell and eventually got stopped out with large losses. Had they held on they probably would have been OK. Here I don’t buy if the absolute price difference between today’s close and yesterday’s is greater than 2 X the ten day average true range. Stay away from too much “VOL.”
Once a trade is put on I use the following logic to keep track of consecutive closing relationships:
Here I am using the IFF function to compare today’s close with the prior day’s. iCnt is a loop counter that goes from 0 to 1. IFF checks the comparison and if it’s true it returns the first value after the comparison and if false it returns the last value. Here if I have two consecutive up closes value1 accumulates to 2. If I am long and I have two up closes I get out. With this template you can easily change this by modifying the input: consecUpClose. Trade management also includes a protective stop and a time based exit. If six days transpire without two up closes then the system gets out – if the market can’t muster two positive closes, then its probably not going to go anywhere. The thing with mean reversion, more so with other types of systems, is the use or non use of a protective stop. Wide stops are really best, because you are betting on the market to revert. Look at the discrepancy of results using different stop levels on this system:
Here an $1,800 stop only cut the max draw down by $1,575. But it came at a cost of $17K in profit. Stops, in the case of Mean Reversion, are really used for the comfort of the trader.
This code has the major components necessary to create a complete trading system. Play around with the code and see if you can come up with a better entry mechanism.