To calculate the VaR at the 90% confidence level using the given dataset, we first order the dataset VaR in the ordered dataset:

The worst 10 return observations are: [-20, -18, -18, -17, -15, -14, -12, -10, -7, -3]

Total number of observations: 30

Position = $0.10 \times 30 + 1 = 3 + 1 = 4$th observation

The 4th worst return in the ordered list is:

-20, -18, -18, -17, -15, -14, -12, -10, -7, -3

Therefore, the VaR at the 90% confidence level is -17%. This means that, with 90% confidence, the maximum expected loss will not exceed 17% over the specified time horizon.