Microstructural Impact of Algorithmic Trading on African Equity Markets: A Causal Evaluation of Efficiency, Liquidity, and Stability

Abstract

The study provides a detailed empirical evaluation of the effect of the automated trading otherwise known as algorithmic trading (AT) on market efficiency, liquidity, and steadiness in ten main African equity markets. The study offers an in-depth empirical assessment of the impact of AT on market efficiency, liquidity, and stability in ten major African equity markets. To effectively close a critical gap in the literature on research on frontier-markets, the study applies an integrative approach to methodology, which combines along with intraday market microstructure evidence, causal inference approaches and machine learning practices. The intensity of algorithmic trading was measured using microstructure tools, and the study measures the consequential variables of the which includes return autocorrelation, bid ask spread, market depth and volatility persistence. We adopt quasi-experimental designs that exploit market-structure reforms for causal identification on a methodological level. By using a Finite Mixture Model, the analysis first classifies continent exchanges into Class 1 Deep Markets (42% probability, average capitalization $245.6B) and Class 2 Thin Markets (58% probability, average capitalization $18.7B). It shows that AT intensity is highly concentrated in deep capitalized venues. Using a robust double machine learning framework with 5-fold cross-fitting, we estimate the causal effects of quoting automatically. The estimates indicate that an increase in Message-to-Trade Ratio (MTR) and Cancellation-to-Trade Ratio (CTR) compresses the transaction cost significantly. In particular, the effective spread decreases by 0.156 (p < 0.01) and 0.167 (p < 0.01), respectively, while increasing log market depth by 0.098 and 0.087. AT robustly improves informational efficiency, reducing Kyle’s  by up to 0.021 units. We quantify a distinct structural trade-off: higher AT activity also leads to a statistically significant rise in realized volatility (MTR effect: +0.027; CTR effect: +0.031). Predictive modeling based on Long Short-Term Memory (LSTM) neural networks demonstrates that high AT environments considerably enhance predictability, despite an increase in baseline volatility; improving out-of-sample RMSE forecasting performance by 11.9% and boosting R2 by 18.5%. To sum up, our high-frequency anomaly detection reveals that AT does not change the market shock frequency baseline of 1.92%. However, it provides a significant shock absorber that compresses the anomaly duration by 3.6 minutes, helps the market recover post-shock in 16.6 fewer minutes, and reduces the extreme volatility multipliers from 4.2x to 3.8x. Most prominently, findings indicate that the African Trade (AT) Exchange is essential for market liquidity and resilience, although the share is not equal. The study finds that a common feature of algorithm trading is its capacity to improve informational efficiency, contributing positively to the information quality of prices. It does so by reducing the return autocorrelation and order flow predictability. Having state-dependent dynamics implies that the sensitivity to shocks runs through the rich posterior distribution of shocks for various parameters. To the point, the effects are quite heterogeneous since they are distinctly moderated by the original depth of the market and the structure of the exchange. A customized policy prescription study for African regulators and market actors to illustrate the need to measure visibility, establish robust market structures, and execute conditional reforms in line with the specifics of the market situation and liquidity needed.