Sparse Index Tracking
Replicate a benchmark using only a small subset of its constituents, balancing tracking accuracy against portfolio cardinality through a reweighted L1 penalty followed by a constrained refit.
Overview
Holding every constituent of a benchmark like Nifty 500 is rarely practical — transaction costs, lot sizes, and operational complexity all scale with the number of names. Sparse Index Tracking finds a small subset of assets whose linear combination most closely tracks the benchmark return time series, then renormalises the active set to a long-only portfolio.
Folio Lab uses an iterative reweighted L1 procedure to identify the sparse support, then a constrained least-squares refit to assign optimal weights within that support.
Algorithm
Step 1 — Reweighted L1 (support discovery)
Iterate the following problem for a small number of cycles. With previous weights and adaptive penalties :
The reweighting concentrates the L1 penalty on small weights and relaxes it on large ones, mimicking the L0 (cardinality) penalty while staying convex at each iteration. by default and .
Step 2 — Cardinality cap
After convergence, the top- weights are kept, where the default cap is (configurable). Names below the threshold are dropped to enforce true sparsity rather than soft shrinkage.
Step 3 — Constrained refit
On the surviving active set, solve a clean long-only least-squares problem to optimise the weights given the chosen support:
This separation of support discovery and weight refitting is a standard trick for sparse problems, avoiding the bias that comes from a single L1 stage.
Advantages & Limitations
Advantages
- Practical: Replicates with a tradeable, small portfolio.
- Two-stage: Avoids L1 shrinkage bias on retained weights.
- Convex sub-problems: Each iteration is a tractable QP.
- Configurable cardinality: Tune sparsity to operational constraints.
Limitations
- Requires benchmark returns: Cannot operate without a target series.
- Local optimum: Reweighted L1 is heuristic vs exact L0.
- Tracking error: Smaller cardinality means worse out-of-sample tracking.
- Sample dependence: Support choice can be unstable across windows.
References
- Candès, E. J., Wakin, M. B., & Boyd, S. P. (2008). "Enhancing sparsity by reweighted L1 minimization." Journal of Fourier Analysis and Applications, 14(5), 877-905.
- Benidis, K., Feng, Y., & Palomar, D. P. (2018). "Sparse portfolios for high-dimensional financial index tracking." IEEE Transactions on Signal Processing, 66(1), 155-170.
- Brodie, J., Daubechies, I., De Mol, C., Giannone, D., & Loris, I. (2009). "Sparse and stable Markowitz portfolios." PNAS, 106(30), 12267-12272.