Data science, a field of study that has stolen the limelight recently, has triggered a massive race in the quantitative trading world. Dhaka-based firm Quantformatix, which provides consultancy and analytical solutions to corporate clients using Quantitative Methods, put a few hypotheses to the test to see whether there are consistent behavioural patterns on listed stocks in the Dhaka Stock Exchange (DSE). While most people in the financial community apply their knowledge of technical analysis on charts to derive ideas of trade, quant strategists rely on machine learning and other statistical methods to build systems that generate trade ideas. A chart can be described in many more ways subjectively than it can be done objectively, and for a quantitative trading model, which is far less ambiguous in its approach, the charts are read as time series variables. Without going into the nitty-gritty, the main purpose of Quantformatix's exercise was to see, if within the constraints of Long only, T+2 and the liquidity gaps, there can be a model that can successfully beat the general Index.
The confusions that surround the understanding of Systematic Trading, Black Box, and HFTs can still be found in abundance. After several types of analysis it was confirmed that Black Box and HFT models are non-applicable for Dhaka Stock Market even to this day. However, the systematic trading picture was different.
Algorithm AD26 happens to harvest alpha from the regime switching characteristics found within market microstructure augmented into a timing model. Like other emerging markets the Dhaka stock market also responded positively for AD26. Although the performance seen was index beating, the question remains as to how scalable this model can be, with increasing size of AUM, and when there are random pockets of liquidity gaps even in some popular stocks. The solution, to some extent, is in the allocation technique that becomes a test on the percentage of the average daily volume smoothed over a period to define the maximum position limits per stock. Unlike discretionally managed funds, the systematic trading model has to have a large set of stocks in its investible universe. The purpose is to keep the power of probability in its favour, thereby allocating nothing more than 3.0 per cent of the fund per stock, given the constraints mentioned. This case, however, can be different for algorithms that are looking at an alternative spectrum of time series for alpha generation, so the allocation rule doesn't have to be concrete.
The quality of an algorithm can be assessed by a back-testing model. It is a model rather than a platform because there are good and bad ones in existence. A back-testing framework built to assess a systematic strategy has to ensure there are no cherry picking in stock selection or data snooping in the process i.e. the test has to run on a walk forward basis from a historical point and have to be transaction-oriented rather than price-centric. A systematic strategy has to have minimum filters to create the investible universe. One of the filters can be average daily volume or value traded as a percentage of market cap. There also has to be a mechanism in places to classify stock time series into a minimal number of price path-dependent categories. This classification will highlight whether the natural behaviour of price due to investors participation level flows in a way that can be predicted by some probabilistic measures. The key metrics that define the breed of a systematic trading model are Sharpe Ratio, maximum drawdowns, win-loss ratio, and the probability of max losing streak under the umbrella of the returns.
With the absence of insider knowledge or the so-called luck factor in a trading model, there is widespread scepticism that something like pure systematic trading is possible with the stocks listed in Bangladesh's exchanges. The reason for this is clearly visible from the relative performance seen between the Mutual Funds and the Index. From the outset, if well-known Mutual Funds are unable to produce index-beating performance then how is it possible by some Algorithm? Well, the answer is quite the contrary. Mutual Funds are built with an investment policy and focus on the forward-looking guidance of a fundamental story. Their target is to achieve maximum return while keeping to a theme that can be scalable as well as be explained under economic terms. These terms, on one hand, provide the unique selling point and perhaps also a burden on the other.
Can the performance numbers be better than currently seen for the Mutual Funds? The study results have shown that invoking a timing model to the Mutual Funds would certainly improve the Mutual Funds performance relative to the Index. A market timed well by one is a missed opportunity for another, so even though there are some room for improvements it will only be possible for a few. All systematic model tested through liquidity constraints has a maximum fund capacity. Scaling up the AUM for a pure systematic trading model that reinvest profits, would eat into the annualised returns year on year and dampen the performance closer to the index. After all, we are operating on a market sized just over $40bn only.
So how hard is it to follow a systematic rule in trading? The shift from discretionary to systematic trading can be an enormous hurdle. The battle is between the core belief of a trader's perception of a stock's expected performance versus the machines that only decide trades on basis of falsification of an expectation that gets recalibrated daily. The portfolio model in illustration covered 86 stocks within pre-defined market cap bracket that had 937 trades over just 2.5 years with a capacity tested up to 500 million (50 crore) without experiencing any return decay. Clearly this is not a massive advantage but certainly a start. All in all, it gets to show where the Dhaka Stock Market resides from a systematic standpoint and what can be achieved.
The writer is a co-founder of Quantformatix and a managing partner at Blue Birch Capital Advisor LLC.