Difference between revisions of "Document:MIT-MFin-essay"

For decades, proponents of the random walk hypothesis have asserted that the market as unpredictable, while Black attributed price movements two factors: random noise trades, and more informed trades which move market price towards the unobservable intrinsic "shadow" price. More recently, evidence was presented (including by Professor Lo) that goes against the random walk hypothesis. At the same time, Mandelbrot has criticized the assumptions of portfolio theory, and proposed using fractals for modelling instead. Given the importance of the financial system, it begs the question of just who is "right", or more accurately, which models and assumptions are closest to real behaviour?

Having just finished an undergraduate degree in MEng Computer Science with a focus on machine learning, most of my experience is with (un)supervised learning, data mining, and working with various forms of data, such as search or advertising data, ontologies, images, DNA sequences etc. Although much of the technical literature in finance is beyond my grasp, I do have qualitative appreciation for the ideas, even if it's coloured by an engineer's perspective. For instance, Markowitz elegantly reduced the selection of portfolios of risky assets to a pareto-optimization problem, the computation of which Sharpe simplified (approximately) by computing a common index, which I liken to how log ratios of gene expression in microarrays are compared among multiple samples. Other than the theory, I am interested to learn about more practical problems like filtering out errors that arise in market tick data, and how the errors should affect our assumptions.

While I am keeping an open view about what I intend to do until I get a better feel for the nature of the work, I tend to gravitate towards either academic or industrial research, which is what I have been doing so far. Either as part of the MFin electives or further studies, I am interested in machine learning methods suitable for financial data, or more generally, for non-independent time series data and random processes. I'm also eager to explore the role of general computational models such as artificial neural networks in improving the robustness of conventional models against situations when their assumptions break down. This concept is not new; Netflix, for instance, runs a piece of software called Chaos Monkey that randomly disables their Amazon AWS server instances, to let engineers simulate and respond to unexpected failures before they naturally happen. However, first I must understand mathematically the advantages and limitations of state-of-the-art methods and concepts in finance, and that is what the core of this programme is about.

Up to now, I have been involved in small-scale software development and the occasional hackathon or two, while on the research side, I have been solving problems specific to the tech industry. I have mostly worked in teams of 2 to 6 on the development front, often with flexible team dynamics as required by agile development methods. Small teams afforded flexibility, but also demanded self-sufficiency and proficiency in multiple roles, so it was a great learning opportunity. The hackathon format was essentially the same, other than the intense time pressure and severe trade-offs one has to consider to be successful. On the other side, the research I was involved in were self-directed projects with input from my colleagues or supervisors, lasting at least 2 to 3 months long, including literature review and implementation. Research decisions had to be defended on a regular basis, especially in industrial settings where the end-goals change and the deadlines are tight. These experiences, as well as my familiarity with experimental overhead like cleaning data, collecting ground truth labels, or computing inter-rater agreement, gives me the tools and practical experience to tackle multifaceted problems confidently.

I share MIT's belief in lifelong learning, and I firmly believe that having an understanding in multiple disciplines provides the necessary insight and expertise to comprehend and ultimately solve today's most difficult problems. I seek out challenging environments, new perspectives and ideas, to change the way I think; that is when I make the most important mistakes, learn the most useful lessons, and develop fastest both personally and professionally, and so I actively work towards putting myself in such situations every opportunity I get. I think the combination of both MIT and the Master of Finance programme is the best fit at this stage in my career, and I look forward to learning and contributing with the very best this summer.