US Sector Climber

The U.S. Sector Climber quantitative investment strategy (the Strategy) invests into 11 U.S. sector ETFs, and a gold ETF. To allocate between these investments, the Strategy performs a mean-variance optimization (MVO) at several points during the day using short term expected returns and volatilities, and longer term correlation estimates. Resulting weights are smoothed before they are translated into share numbers.

1. Expected Return and Covariance Estimation

For each asset, returns are estimated using exponentially decaying log returns of VWAP prices over a 10 minute window, every 10 minutes during the trading day which typically starts at 9:30am and ends at 4pm. To avoid fitting to outliers, two averages are used; the decay factors ${\lambda }_i$ are set to ${\lambda }_1=0.97$ and ${\lambda }_2=0.94$. The two averages are then combined using weights proportional to $\frac{1}{{\sigma }_i}$ where ${\sigma }_i$ is the standard deviation of the return, calculated using the same decay factors.

To calculate the covariance matrix we first estimate an exponentially decaying covariance matrix and back out the correlation matrix. We then use a slow decay factor of ${\lambda }_{\mathrm{cov}}=0.99987148154578$ to update the previous correlation matrix with the new one. In a second step we estimate variances of the returns using a fast decay factor of ${\lambda }_{\mathrm{var}}=0.97$ and calculate the final covariance matrix using the correlation matrix and the variance.

This approach is more robust as correlations are typically slower moving, while variances can change quickly. Moreover, by separating correlations and variances we can more easily deal with missing values.

2. Optimization

At specified points in time throughout the trading day the Strategy runs a MVO using the covariances and expected returns as above. The MVO aims to find a fully invested portfolio with the highest return for a 10% volatility target. The result of the MVO is weighted against the previous weights using ${\alpha }_{\mathrm{MVO}}=0.1$ for the new target weights.

3. Volatility Targeting

Every 10 minutes the Strategy checks its ex-ante volatility using the covariance matrix estimated above. If the ex-ante volatility is larger than 10%, it will scale down all weights to reach an estimated volatility of 10%. If volatility is estimated to be below 10% it will scale up weights up to a maximum total of 1.0: the strategy leverage should never exceed 1.

Volatility targeting happens after the MVO in case a MVO is performed at the same instant.

4. Smoothing

To avoid high turnover, weight changes have to be at least 2% of the total Strategy level. Any change to weights below that threshold will be ignored, and the old value is used instead.

5. Implementation and Trading

The resulting weights are translated into share quantities using the previous strategy level, calculated from last trading prices. If current trading prices are missing, the last known value is used. After share quantities are calculated, the central hedging facility is notified of the change, at which point it will update client portfolios by trading to the new quantities.

Strategy construction as of 3/30/2023. To adapt to structural changes in the market, GLR Technologies may tweak certain parameters at any time. If the situation allows, GLR Technologies will inform clients ahead of any changes. For a major update GLR will typically create a new version of the strategy, at which point clients can switch their allocations to the new version if desired.