Marco - I seem to recall that you indicated that variable position sizing would be released “next week”. That was a couple of weeks ago. What is the status of that? This is a big feature that I and I’m sure others have been waiting for a long time.
Thanks
Steve
In progress. It’s close.
Out of curiosity… how would You like it to control the sizes ?
Marco,
I would be interesting as to whether giving a slightly higher weight to the stocks with higher rank position would benefit a port.
So if 2 stocks were recommended and one stock had RankPos #1 and the other recommended stock was RankPos #5 (RankPos #2, 3 and 4 already held) would giving the stock ranked #1 a greater weight (maybe 10% more money) help my port/sim?
Thanks.
-Jim
Marco, I sent you a white paper on the subject a few years ago.
I would like to be able to size positions by market cap, inverse volatility, inverse correlation to other holdings, or by rank.
(Shares purchased/Daily volume)^0.6 is linearly related to the slippage.
It would be tempting to set a maximum for “Shares purchased/Expected daily volume” but the stocks with less daily volume also have higher returns.
So, one would rationally develop an equation: Expected Returns = Function (multiple factors including expected daily volume and perhaps rank etc) - (Shares purchased/Expected daily volume)^0.6. Daily volume is not PIT, of course, but the 5 day average might be an adequate surrogate.
And weigh the purchases based on this function of expected returns.
In any case, the ability to use a function would ultimately be helpful. In truth, I am not quite to the point where I would change a purchase based on the liquidity of a stock. But I am getting there.
I am very sure other members have better ideas for functions of expected returns.
I suspect this is what is ultimately planned—at least long term if not immediately. Personally, I will be happy with any incremental improvement including just factor weights of a few limited factors (e.g., RankPos).
Thanks.
-Jim
Marco:
I would like to see (test) how sizing positions based on ATRN would affect various metrics including return and drawdown. Thanks.
Bill
My desire is to have the following:
Port / Designer Model:
(1) Select normalization type: #All, #Sector, #Industry (The sum of all weights for the stock positions adds up to either 100%, SecWeight or IndWeight.)
(2) Select normalization (or weight mapping) method: constant (equal weight), linear, exponential, logarithmic
(3) Specify number of positions per (1) - number of positions per the universe, sector or industry.
(4) Specify stock weighting formula (including stock fundamentals, technicals, custom series, etc). The weighting formula is normalized per (2) and applied to (1) - the universe, sector or industry.
(5) Specify a “rebalance” threshold. Only rebalance the positions if the total error of positions (all, sector or industry) is greater than x% from ideal
(6) eliminate insignificant %weights / superficial buy/sell % differences
Since this is a mouthful for newbies and less experienced users, there should be some predefined (wizard) settings that can be called up such as inverse volatility.
Books / Designer Model Books:
(2) Select port normalization method (100% or 100%+ if trading on margin)
(3) Select number of ports
(4) Specify port weighting formula
(5) Specify port rebalance threshold
(6) eliminate insignificant stock %weights / superficial buy/sell % differences
NOTE: Books need to flow through the individual port rebalancing then calculate the book weighting/rebalancing
We’ll be supporting (4) and (6) for portfolios to start.
But don’t worry, further enhancements are planned for this, as we consider the ability to rebalance without reconstituting a critical improvement to managing investments with P123.
That’s great news. I hope it will work for ETFs with formulas based on momentum, volatility, macro data, and if possible any custom data series.
Thanks Marco and Aaron!
Back in my hedge fund days, we used to size positions based on their “Z-scores”. Each stock would be traded long or short as a proportion of the total portfolio based on some proprietary metric. This would be similar to a rank-weighting in P123. I would like this method because it would allow users to define their own bet sizes based on some “optimal” expectancy.
If nothing else, market cap sizing would allow models to be more directly comparable with cap-weighted indices… No more wondering if a model is simply milking the size anomaly.
Position size as a function of Regular Dividends would be nice.
Volatility based measures make the most sense to me.
Marco - when is Variable Position Sizing coming?
Some last kinks are still being worked out in the algorithm to ensure this feature is launched with reasonable and robust behavior. Once that’s done, it’ll be ready to be used exclusively in portfolio backtests. Work has not yet commenced on integrating it into the live portfolio reconstitution/rebalance workflow, so that’ll be the next phase for this project.
cool