SignalTrace

Markov-switching Asset Allocation: Do Profitable Strategies Exist?

John Montgomery4 min read

Bulla et al. 2010

“Markov-switching Asset Allocation: Do Profitable Strategies Exist?” by Jan Bulla, Sascha Mergner, Ingo Bulla, André Sesboüé and Christophe Chesneau circulated as MPRA Paper 21154 (January 2010) and was published in 2011 in the Journal of Asset Management (vol. 12, issue 5, pp. 310-321). It is the vault’s most disciplined positive evidence for a tradeable Markov Regime-Switching Model application. The authors fit a two-state Gaussian Markov-switching model to daily returns of five major equity indices — the DAX, DJIA, NASDAQ 100, Nikkei 225 and S&P 500 — over roughly four decades. Crucially the two states are defined purely by volatility: a low-variance regime and a high-variance regime. The strategy tested is the simple 1” timing rule that later became a literature benchmark: hold 100% equity when the low-variance regime is forecast, hold 100% cash otherwise.

What separates this paper from weaker backtest claims is its explicit attention to realistic out-of-sample conditions. Estimation uses a rolling 2000-day window; the next-day regime is forecast by Viterbi-decoding the most probable state path; and a k=6 median filter smooths the signal specifically to suppress short-lived flips that would otherwise generate excessive turnover. One-way transaction costs are charged at 10 basis points (0.10%) — a level that has since become the standard daily-trading cost assumption in the regime-switching literature. Under these conditions the strategy is profitable after costs for all five indices: portfolio volatility falls by an average of about 41%, out-of-sample Sharpe ratios run 0.437-0.646 against an index average of 0.342, and annualised excess return over buy-and-hold ranges from 18.5bp on the S&P 500 to 201.6bp on the Nikkei.

The honest reading of those numbers is that the edge is overwhelmingly risk reduction, not return generation. An 18.5bp annual excess return is economically trivial and well within the noise of cost and slippage assumptions; the headline benefit is that the strategy sidesteps high-volatility (and historically falling) markets, which mechanically lifts the Sharpe ratio because high-volatility periods tend to coincide with declines (the Schwert effect). The authors themselves stress that out-of-sample excess returns are “naturally much lower than in-sample,” that misclassification jumps to roughly 10% at the start and end of each rolling window, and that a single wrong regime call can be “detrimental” rather than merely sub-optimal. The median filter exists precisely because, without it, transaction costs eat the edge — an admission that the strategy lives close to the cost frontier.

This paper sits at the centre of the vault’s empirical-trading cluster. It is the benchmark model that later Statistical Jump Model and asset-specific regime studies (e.g. Shu, Yu and Mulvey) compare against, and the “0/1 strategy” they all adopt originates here. Read alongside Dacco and Satchell 1999, the relationship is instructive: Dacco & Satchell prove analytically that regime-misclassification destroys forecasting advantage, and Bulla et al.’s design — daily data so a wrong call costs one day not one month, Viterbi paths, median filtering — is essentially an engineering response to that theorem. The result substantiates regime-switching as a defensible risk-management overlay but not as a directional alpha engine. No public code or data release accompanies the paper, so independent replication has not been formally established, though the 0/1 strategy has been re-implemented many times by follow-on studies.

Profitability grade — moderate. The study clears most bars: genuine out-of-sample testing, explicit transaction costs, a buy-and-hold benchmark, and a risk metric. It falls short of strong because the after-cost excess return is tiny, slippage and market impact are not modelled, no formal robustness battery (e.g. Deflated Sharpe Ratio or purged cross-validation) is reported, and there is no independent replication or code release. The profitability that exists is real but marginal and is better described as volatility reduction than as alpha.

Bulla et al. 2010 [supports] Regime-Based Asset Allocation Bulla et al. 2010 [tests-strategy] Regime-Based Asset Allocation Markov Regime-Switching Model [precedes] Statistical Jump Model Dacco and Satchell 1999 [relates] Bulla et al. 2010

Connections

Sources

Graph View

Backlinks