Love the emphasis on covariance being more critical than mean estimation for portfolio performance. The cross-geo correlation data (0.08 US-Canada, 0.10 Canada-Australia) is surprising low and creates real diversification value. That jump from 1.66 to 1.76 Sharpe with MVO shows proper allocation matters even when you're only dealing with a few sleeves instea of hundreds of individual names.
Hi, thanks for the interesting post. What does it mean that the equal weight is rebalnced annually?
If I understand it correctly, it means that the actual weights change throught the year while for the mean-variance portfolio the weights are constant for a year which mean that everyday you have to rebalance your portfolio daily in order to keep the weights constant. Thus, in the comparison between the 2 portfolios in reality there is a drag in the mean-variance one due to daily rebalance, correct?
Thank you for the clear and educational post-it is very helpful. From a practitioner's perspective, I would like to add a few implementation-related considerations. First, because the strategy relies on index constituents, the backtest should use historical index membership at each rebalance date; using 2025 constituents to rebalance a 2015 portfolio introduces survivorship bias, as underperforming names may have already been removed, so the investable universe at each rebalance should reflect only the stocks that were index members at that time. Second, the strategy appears to rely on a static end-of-year (December) annual rebalancing choice, which may introduce timing bias or hidden seasonality; to improve robustness and reduce potential data-snooping concerns, it would be useful to test the same strategy using month-end rebalancing dates and compare results across months. Finally, I would be interested in understanding the transaction cost assumptions used in the backtest, as well as the average and distribution of portfolio turnover, since these factors often dominate real-world performance.
Yep, the backtest uses point-in-time index membership (via Norgate), so the investable universe at each rebalance date is the actual historical constituent set (no survivorship). Not doing that would be a rookie mistake.
On costs: I assume 2 bps per trade (well above commissions). And with MOC execution in highly liquid names, slippage shouldn’t be a limiting factor unless you’re trading truly massive size (billions.... which I’m assuming isn’t the case here).
Love the emphasis on covariance being more critical than mean estimation for portfolio performance. The cross-geo correlation data (0.08 US-Canada, 0.10 Canada-Australia) is surprising low and creates real diversification value. That jump from 1.66 to 1.76 Sharpe with MVO shows proper allocation matters even when you're only dealing with a few sleeves instea of hundreds of individual names.
Hi, thanks for the interesting post. What does it mean that the equal weight is rebalnced annually?
If I understand it correctly, it means that the actual weights change throught the year while for the mean-variance portfolio the weights are constant for a year which mean that everyday you have to rebalance your portfolio daily in order to keep the weights constant. Thus, in the comparison between the 2 portfolios in reality there is a drag in the mean-variance one due to daily rebalance, correct?
Thank you for the clear and educational post-it is very helpful. From a practitioner's perspective, I would like to add a few implementation-related considerations. First, because the strategy relies on index constituents, the backtest should use historical index membership at each rebalance date; using 2025 constituents to rebalance a 2015 portfolio introduces survivorship bias, as underperforming names may have already been removed, so the investable universe at each rebalance should reflect only the stocks that were index members at that time. Second, the strategy appears to rely on a static end-of-year (December) annual rebalancing choice, which may introduce timing bias or hidden seasonality; to improve robustness and reduce potential data-snooping concerns, it would be useful to test the same strategy using month-end rebalancing dates and compare results across months. Finally, I would be interested in understanding the transaction cost assumptions used in the backtest, as well as the average and distribution of portfolio turnover, since these factors often dominate real-world performance.
Yep, the backtest uses point-in-time index membership (via Norgate), so the investable universe at each rebalance date is the actual historical constituent set (no survivorship). Not doing that would be a rookie mistake.
On costs: I assume 2 bps per trade (well above commissions). And with MOC execution in highly liquid names, slippage shouldn’t be a limiting factor unless you’re trading truly massive size (billions.... which I’m assuming isn’t the case here).