Based on our understanding, while Algo-Trading may NOT be a bullet-proof technique, its ability to process a huge set of data coupled with its systematic approach and low involvement of emotion, making it an approach that based on calculated risk so long as exit rules and stop-loss level are probably defined. In our view, this technique may even be more useful in a chaotic market condition when all trading rules were preset. As of end-Oct 2018, out of the 5 Models (using monthly data) that we designed, the RSI Model shows a B.O.W. signal due to oversold technical condition, but the Mean-Reverting Historical PER Band Model is still emanating S.O.S. signal which was shown since end-Jul 2018. The other 3 models have yet to give any trading signals.

The Workshop. Of late, we have engaged QuantInsti, an algorithmic trading research and training institute, to conduct a full-day workshop on algorithmic trading (“Algo-Trading”). The course was wellattended with approximately 70 personnel from both internal and external parties. The workshop aims to help the participants to gain basic understanding of Algo-Trading, including quantitative strategy modelling, back-testing, risk management, system infrastructure and technology involved as well as industry trend and practice.

What is Algo-Trading? In a nutshell, Algorithmic Trading (algo-trading / automated trading / black-box trading) is the process of using computers programed to follow a defined set of instructions (an algorithm) for placing a trade at a speed and frequency that is impossible for a human trader (see Figure 1). The defined sets of rules are based on timing, price, quantity or any quantitative models. Apart from profit opportunities for the trader, algo-trading makes markets more liquid and renders trading more systematic by ruling out the impact of human emotion and error in trading activities. We understand that Algo-Trading accounted for ~40-50% market share of Indian stock market currently.

Systematic vs Discretionary. Note that the terms ‘quantitative’, ‘systematic’ and ‘rules-based’ are often used interchangeably; they represent an investment approach that is often perceived to be in direct opposition to a ‘fundamental’, ‘discretionary’ or ‘stock-picking’ approach, which is NOT true. While it is fair to contrast systematic and discretionary approaches, investors should be aware that these ideas are not contradictory. Indeed, both systematic and discretionary managers pursue the same objective and both can be fundamentally-driven judging from the very similar inputs. The difference is probably the ways or approaches, which, again share the same and singular goal of improving investment performance. Empirical studies show that neither systematic nor discretionary managers are inherently superior. Each has the potential to deliver good investment outcomes and there is little evidence to suggest that one approach is better than the other. The historical correlations between excess returns from systematic and discretionary managers are low, which suggests that many investors may benefit from incorporating both types into their allocations. Likewise, historical correlations among systematic investors are also low, as low as they are among discretionary investors, suggesting that the notion that "all quants trade on the same signals" is misplaced. Nonetheless, we do reckon that discretionary approach can potentially offer superior upside due to deep and proprietary insights. Systematic approach, on the other hand, is able to offer superior consistency and superior information processing.

Back-testing Various Models. During the workshop, a “breakout” strategy and a “Pair Trading” strategy were set as examples. From Figure 2 – The Spectrum of Strategy – these two strategies are commonly and widely adopted quantitative strategies that based on Trend (or Momentum) and Mean-Reverting concepts. Note that the various strategies shown in Figure 2 are also known as Systematic Fundamental. Based on the similar concept, we can now use it to back test a few of our Timing Models (see Figure 3-7). However, do note that back-testing is useful for hypothesis testing BUT not a data-fitting tool. Data-mining may generate good performance based on historical records but can be bad going forward. Hence, the key word in back and forward testings is “consistency”. As a rule of thumb, a strategy with low Sharpe ratio, concentrated wins (or low stability or low hit rate), large downside risks (negative skew or unfavourable reward-to-risk ratio) and large drawdowns is not a good system. Portfolio Optimisation. One also note that, for portfolio optimization, Kelly’s Creterion is widely adopted instead of Markowitz mean-variance portfolio optimization. The Kelly's Criterion is well-known among gamblers and investors as a method for maximizing the returns (or long-term capital growth).

The formula of Kelly’s Optimal Fixed Fraction = {[(Average gain on +ve Trades / Average Loss on –ve Trades)+1] * (Winning Probability or Hit Rate)} – 1 / (Average gain on +ve Trades / Average Loss on –ve Trades)

For instance, based on our back-testing results of Mean-Reverting Model 1, investors should allocate 40.1% of the allocated fund for this trading technique (see Figure 3-7).

Algo Specific Risks. The advantage of in Algo-Trading – without human input – is also its major risk. Imagine in a trading environment where orders are sent and executed without human control in a fast pace manner, tremendous damage could be done before a human can realize and respond. In a nutshell, these risks can be classified into the following categories: (i) Access, (ii) Consistency, (iii) Quality, (iv) Algorithm, (v) Technology, and (vi) Scalability. These risks have to be handled within the applications, and before generating an order in the Order Management System. Moreover, it is pertinent that the trader understands the internal working of the black-box (see Figure 8-11).

Source: Kenanga Research - 7 Nov 2018