<p>We all are living in a digital era. The more the teams at Intel, AMD, AWS, and Microsoft work hard and innovate on processing power/flexibilities, the more room it opens up for the world to explore new domains of compute.<br /><br />Cloud computing has been a big game changer for the compute world where there doesn’t exist a need to provision or buy servers anymore for test activities or production. Capex has converted to opex and this has helped us leverage or try out new ideas which were otherwise not feasible due to high risks of failure at the research level.<br /><br />If we look at a regular day for many job profiles and think can computers help, I can find at least two common reasons which pulls back the thought of most namely — Subjectivity and Complexity. For example, we get a portfolio evaluated by a Portfolio Manager because we feel that his experience can help in better diversification and allocation, or for instance, we rely on a Technical Analyst for a forecast on an instrument. Let’s drill down a bit to understand what these professional minds try to do.<br /><br />The advantage that they have over others is more data in brain, with the experience that they hold, as they have lived the moment of those patterns being formed, and the result of them or, the stocks/sectors being volatile or, the predictability of economic indicators. Even though our brain has an estimate to compute this mathematical problem, it was always thought difficult to step in that grey area and think of moving that compute to an Intel processor.</p>.<p>The brain stores data just like any database and when requested, processes that information similar to what a processor does. But the question we need to ask is do we get the optimal output? It’s more of a ‘feel’ than a probability analysis and this is what people call <br />subjectivity. To me, subjectivity is just an unexplored domain and therein lies the opportunity for us.<br /><br />The assumption on which we all stand today in the financial markets as forecasters is because we believe that the price points have a Time Series Correlation (very much proven by many thesis), and that is what makes forecasts possible.<br /><br />Today, we stand in a world where subjectivity can be dealt with processes like Machine Learning, Neural Network, etc. One of the important reasons for forecasts failing consistently is not having a probability number associated with it. Human behaviour has been a constant since the time humans have existed. If we believe that having a car is a symbol of status, then it always existed in history in the form of horses.<br /><br /> I like Technical Analysis a lot not because you have oscillators or indicators to help you out, but it’s more about understanding how humans behave. We don’t really need to live 1,000 years to be the most experienced. As long as data is available, we can study it well and derive patterns from it. Life is very small to make all mistakes and learn from it, instead we can learn it from others’ mistakes happening around or looking at historical mistakes.<br /><br />How can these things help us? There are multiple utilities that can be created around these, and two areas of implementation could be research and advisory.<br /><br />Research: As we have been discussing the research process in the earlier examples, we can break down human behaviour as rules and delegate them to computers which is much faster and accurate than our brain. It can solve complex problems like finding you an optimal allocation from millions of correlations and diversification possibilities.<br /><br />An as simple as choosing 20 stocks from Nifty 100 Index leaving apart constraints has 53,59,83,370 trillion possibilities to be precise and our brain doesn’t have the right optimisation algorithm to choose one from that. If we are able to solve this problem, what we achieve is an incremental probability of returns to our portfolio. Now, a more complex problem is prices changing each second, how we know when to make the next change? Today, when compute of such complex problems can be handled it’s we who need to leverage it.<br /><br />We often hear different Technical Analysts coming up with different forecasts on the grounds of the same indicator for eg. RSI. Now, how is this possible? The problem is again trillions of possibilities due to the variables that go within. An optimal solution to this problem might be a probability distribution of the outcomes for a pattern identification which will almost define the chance of it working.<br /><br />Advisory: What do we expect from an advisor on the trading floor? Recommendations coming in from research, tracking of positions, other back office activities, etc. Could we not automate these and reach all in real time? Doing this would make advisory relevant. An advisor also deals with the same problem of too much data, as his brain might not be capable enough to compute from every tick as to how and which client may get impacted and on the other hand a smart algorithm may never miss an opportunity on parameters we set. Tracking can become robust enough to generate desired events.<br /><br />If we expect automated cars to reach the road soon, I’m sure the buildings around those roads won’t remain without technological innovation. Robo advisory may sound fancy today, but will be a basic need tomorrow. To fight a gun, a sword might not suffice.<br /><br />(The writer is head of Quantitative Research at MOSL)</p>
<p>We all are living in a digital era. The more the teams at Intel, AMD, AWS, and Microsoft work hard and innovate on processing power/flexibilities, the more room it opens up for the world to explore new domains of compute.<br /><br />Cloud computing has been a big game changer for the compute world where there doesn’t exist a need to provision or buy servers anymore for test activities or production. Capex has converted to opex and this has helped us leverage or try out new ideas which were otherwise not feasible due to high risks of failure at the research level.<br /><br />If we look at a regular day for many job profiles and think can computers help, I can find at least two common reasons which pulls back the thought of most namely — Subjectivity and Complexity. For example, we get a portfolio evaluated by a Portfolio Manager because we feel that his experience can help in better diversification and allocation, or for instance, we rely on a Technical Analyst for a forecast on an instrument. Let’s drill down a bit to understand what these professional minds try to do.<br /><br />The advantage that they have over others is more data in brain, with the experience that they hold, as they have lived the moment of those patterns being formed, and the result of them or, the stocks/sectors being volatile or, the predictability of economic indicators. Even though our brain has an estimate to compute this mathematical problem, it was always thought difficult to step in that grey area and think of moving that compute to an Intel processor.</p>.<p>The brain stores data just like any database and when requested, processes that information similar to what a processor does. But the question we need to ask is do we get the optimal output? It’s more of a ‘feel’ than a probability analysis and this is what people call <br />subjectivity. To me, subjectivity is just an unexplored domain and therein lies the opportunity for us.<br /><br />The assumption on which we all stand today in the financial markets as forecasters is because we believe that the price points have a Time Series Correlation (very much proven by many thesis), and that is what makes forecasts possible.<br /><br />Today, we stand in a world where subjectivity can be dealt with processes like Machine Learning, Neural Network, etc. One of the important reasons for forecasts failing consistently is not having a probability number associated with it. Human behaviour has been a constant since the time humans have existed. If we believe that having a car is a symbol of status, then it always existed in history in the form of horses.<br /><br /> I like Technical Analysis a lot not because you have oscillators or indicators to help you out, but it’s more about understanding how humans behave. We don’t really need to live 1,000 years to be the most experienced. As long as data is available, we can study it well and derive patterns from it. Life is very small to make all mistakes and learn from it, instead we can learn it from others’ mistakes happening around or looking at historical mistakes.<br /><br />How can these things help us? There are multiple utilities that can be created around these, and two areas of implementation could be research and advisory.<br /><br />Research: As we have been discussing the research process in the earlier examples, we can break down human behaviour as rules and delegate them to computers which is much faster and accurate than our brain. It can solve complex problems like finding you an optimal allocation from millions of correlations and diversification possibilities.<br /><br />An as simple as choosing 20 stocks from Nifty 100 Index leaving apart constraints has 53,59,83,370 trillion possibilities to be precise and our brain doesn’t have the right optimisation algorithm to choose one from that. If we are able to solve this problem, what we achieve is an incremental probability of returns to our portfolio. Now, a more complex problem is prices changing each second, how we know when to make the next change? Today, when compute of such complex problems can be handled it’s we who need to leverage it.<br /><br />We often hear different Technical Analysts coming up with different forecasts on the grounds of the same indicator for eg. RSI. Now, how is this possible? The problem is again trillions of possibilities due to the variables that go within. An optimal solution to this problem might be a probability distribution of the outcomes for a pattern identification which will almost define the chance of it working.<br /><br />Advisory: What do we expect from an advisor on the trading floor? Recommendations coming in from research, tracking of positions, other back office activities, etc. Could we not automate these and reach all in real time? Doing this would make advisory relevant. An advisor also deals with the same problem of too much data, as his brain might not be capable enough to compute from every tick as to how and which client may get impacted and on the other hand a smart algorithm may never miss an opportunity on parameters we set. Tracking can become robust enough to generate desired events.<br /><br />If we expect automated cars to reach the road soon, I’m sure the buildings around those roads won’t remain without technological innovation. Robo advisory may sound fancy today, but will be a basic need tomorrow. To fight a gun, a sword might not suffice.<br /><br />(The writer is head of Quantitative Research at MOSL)</p>
