Liquidity regimes are an intuitive yet elusive concept. It is easy to conceptualize how markets have periods of high liquidity, where the cost of trading is low, and periods of low liquidity, where a single transaction can alter the market price. However, in contrast to the focus on price for return regimes, there is no single metric we can point to as the proxy for liquidity. To capture the multifaceted nature of market liquidity, this paper proposes a novel framework of extracting insights from scattered liquidity metrics through Gaussian Mixture Model-based Hidden Markov Models (HMMs) to identify historical liquidity regimes. The Gaussian Mixture Model is chosen for its best modelling fit for our time series data, while Hidden Markov Models capitalize on the non-observable nature of liquidity. Training our models with SPDR S\&P 500 ETF trading data, we evaluate the model performance by backtesting trading strategies based on detected regimes, using both in-sample S\&P 500 futures and out-of-sample volatility and commodity instruments. Our contributions are twofold: 1) We demonstrate the efficacy of regime-switching models in extracting liquidity insights conducive to asset allocation. 2) We propose the integration of Principal Component Analysis with the Hidden Markov Model which enhances convergence for a 3-state model previously deemed unstable. Through these results, our paper adds to the ongoing research to incorporate liquidity insights into portfolio management.