Will the uni course be using pysystemtrade? Perhaps you could set them some course work to add some machine learning forecast functions to plug-in


If you don't want to do your own stitching, you should have a look at quandl's db which offers a stitched database of the main futures contracts based on simple algos such as OI crossover.Hi guys,
based on my limited understanding, price stitching of futures seem to involve a lot of factors that is not easily coded into rules.
Do we have to eyeball each individual instrument and look at volume, seasonality etc to decide on roll dates?
How do you do it?
How does quantopian / quantconnect / quandl do it (if they do it) ?
@traider the simplest form of stitching is the so-called Panama stitching method. @globalarbtrader wrote an extensive post on his blog about how to use this.
It is here: https://qoppac.blogspot.com/2015/05/systems-building-futures-rolling.html
Let's say you use the Panama method, you still have to decide when to a contract ends and when the next starts right? Will everyone will pick the same dates?
I made a program that I check every day. It gathers prices for all my instruments, cycling through one at a time. It keeps track of the standard roll schedule of each contract, and when one of those dates approaches, it shows me details of the contract I'm holding versus the next contract. Once the volume and bid/ask spread of the next contract is close to the one I'm holding, I decide to roll, and I manually execute the trade or trades.
I don’t know what everyone does. I myself select a next contract when the current contract approaches its expiry date, or first notice date in case of contracts which have physical delivery. Deciding which contract to use is something which I have not automated.Let's say you use the Panama method, you still have to decide when to a contract ends and when the next starts right? Will everyone will pick the same dates?
Will the uni course be using pysystemtrade? Perhaps you could set them some course work to add some machine learning forecast functions to plug-in![]()
I have a question about portfolio bootstrapping.
To be clear, I'm using past data to find the weights for instruments in a portfolio that generated the highest Sharpe ratio.
The thing is, this can up-weight highly correlated things because they worked over the time sample (e.g. vstoxx/vix), but really we have no expectation of what the future Sharpe might be.
In other words, ought the weights not be inversely proportional to the correlations?
This should minimise volatility, which we can know with reasonable certainty, and we know the correlations with almost absolute certainty, and ignores any expectation of return, which is unknown.
PS all current weights are here: