# Quantitative Finance Collector is a blog on Quantitative finance analysis, financial engineering methods in mathematical finance focusing on derivative pricing, quantitative trading and quantitative risk management. Random thoughts on financial markets and personal staff are posted at the sub personal blog.

Aug
17

It is really not easy to find online video courses on algorithmic and quantitative trading for newbies, until recently I got an email introducing this website,

**QuantInsti**is Asia premiere Algorithmic Trading Research and Training Institute focused on preparing financial market professionals and equipping them to conquer the contemporary field of Algorithmic and High-Frequency Trading. As an initiative by financial market professionals with stellar academic and professional credentials, the program aims to fulfill the pressing demands for highly specialized skill sets of a potentially lucrative domain of Algorithmic Trading.

Some videos are really useful so I list them below:

**Introduction to Machine Learning for Trading**

This course will help you in understanding the basics of Machine Learning and give you an opportunity to code an algorithm in Python. You will also learn the different types of Machine Learning Techniques. In this course, you will understand a few research papers on Reinforcement learning and how it works. Finally, you will code a trading strategy using the predictions made by the Machine Learning algorithm.

**Trading with Machine Learning: Regression**

In this course, you will learn the mathematical concepts behind regression function, such as the gradient descent and cost function optimization. Diving deeper, you will get a thorough overview of the Bias and Variance problems faced by machine learning algorithms. Finally, you will be coding a trading strategy using the predictions made by the algorithm.

**Trading using Options Sentiment Indicators**

This course will help you to understand the two major emotions that drive the entire market - Fear and Greed and how we can capitalize on them to make profits. We will learn about sentiment indicators, how to interpret them and devise trading strategies using the same. We will develop a trading logic for our algorithm defining our entry and exit points using signals from the sentiment indicators. We will code this strategy using Python programming language and analyze the buy/sell orders of S&P 500 index futures, generated by our strategy in Microsoft Excel using 2 years of historical data for backtesting.

**Statistical Arbitrage Trading**

This course will help you in gaining an in-depth understanding of Statistical Arbitrage along with the key statistical concepts involved in modeling a Stat Arb strategy. This course will give you an insight into the various types of risks involved in a Stat Arb strategy and ways to mitigate them so that you are empowered to start trading pairs or develop your very own Statistical Arbitrage model. We will take you through a practical implementation of Pairs Trading (Lead - Aluminium Pair) in Excel and explain how to code the same strategy in Python as well.

Mar
8

I just returned Beijing from the Midwest Finance Association 2016 Annual Meeting in Atlanta, it is my first time in America, and the life there is quite different from that in the British cities... few people in downtown, hard to go out without a car, people are less friendly (at least look like)...

MFA annual conference provides a forum for the interaction of finance academics and practitioners to share scholarly activity and current practice so as to encourage and facilitate the betterment of the profession. Below I select several papers with download links that are of interest to me, it is by no means a list of top quality of the conference though.

paper

paper

MFA annual conference provides a forum for the interaction of finance academics and practitioners to share scholarly activity and current practice so as to encourage and facilitate the betterment of the profession. Below I select several papers with download links that are of interest to me, it is by no means a list of top quality of the conference though.

**Short-Term Trading Skill: An Analysis of Investor Heterogeneity and Execution Quality**: We examine short-horizon return predictability using a unique, proprietary data set across a large universe of institutional traders with known (masked) identity. We propose a model to estimate an investor-specific short-term trading skill and find that there is pronounced heterogeneity in predicting short-term returns among institutional investors. This suggests that short-term information asymmetry is a significant motivation for trade. Our model illustrates that incorporating short-term predictive ability explains a much higher fraction of short-term asset returns and enables more accurate estimation of price impact. A simple trading strategy exploiting our estimates of skill yields statistically significant abnormal return when benchmarked against a four-factor model. We investigate the source of variation in short-term trading skill and find strong evidence that skilled traders are able to predict short-term returns by following a short-term momentum strategy. Furthermore, we illustrate that the variation in short-term trading skill is statistically dependent on order characteristics such as duration and relative size, that are associated with more urgent and more informed trading. Finally, using both trading skill estimates emerging from our model and proposed skill predictive variables, we show that investor heterogeneity has major implications for quantifying execution quality.paper

**An Empirical Detection of HFT Strategies**: This paper detects empirically the presence of High Frequency Trading strategies from public data and examines their impact on financial markets. The objective is to provide a structured and strategic approach to isolate signal from noise in a high frequency setting. In order to prove the suitability of the proposed approach, several HFT strategies are evaluated on the basis of their market impact, performance and main characteristics.paper

Dec
8

Choosing an appropriate performance measure is important for fund investors, nevertheless, many researchers find empirically that the choice of measures does not matter because those measures generate identical rank ordering, even though the distribution of fund returns is non-normal. In this paper we certify their findings by proving the monotonicity of several widely used performance measures when the distribution is a location-scale family. The mutual fund monthly return data from 1997 to 2015, together with simulation results, collaborate with our proof.

An adequate risk-adjusted return performance measure to select investment funds is crucial for financial analysts and investors. Sharpe ratio has become a standard measure by adjusting the return of a fund by its standard deviation (Sharpe, 1966), nevertheless, practitioners often question this measure mainly for its invalidity if the distribution of fund returns is beyond normal (Kao, 2002; Amin and Kat, 2003; Gregoriou and Gueyie, 2003, Cavenaile, et al, 2011, Di Cesare, et al, 2014). Several new measures have been proposed and investigated to overcome this limitation of the Sharpe ratio, however, Eling (2008)

finds choosing a performance measure is not critical to mutual fund evaluation, Eling and Schuhmacher (2007) compare the Sharpe ratio with 12 other measures for hedge funds and conclude that the Sharpe ratio and other measures generate virtually identical rank ordering, despite the significant deviations from normal distribution. Similar evaluation includes Eling and Faust (2010) on funds in emerging markets, Auer and Schuhmacher (2013) on hedge funds, and Auer (2015) on commodity investments.

This paper proves that several widely used performance measures are monotonic if the distribution of asset returns is a LS family, a family of univariate probability distributions parametrized by a location and a non-negative scale parameters that is commonly applied in finance (Levy and Duchin, 2004). Our proof certifies the empirical findings in other studies on the indifference of choosing a performance measure when valuing a fund. We show that those measures generate virtually the same rank ordering using monthly mutual fund return data from 1997 to 2005 and Monte-Carlo simulations. Therefore this paper contributes to both the academia and industry by clarifying the phenomenon.

For example, the below figure plots the correlation and confidence intervals based on 2000 simulations for each sample size. For simplicity, we show the results for the Sharpe (ρ1), the Sharpe-Omega (ρ2) and the Sortino ratio (ρ3) only. Consistent with the previous finding, the rank correlation among these performance measures is roughly equal, and is approaching one with the increase of sample size.

An adequate risk-adjusted return performance measure to select investment funds is crucial for financial analysts and investors. Sharpe ratio has become a standard measure by adjusting the return of a fund by its standard deviation (Sharpe, 1966), nevertheless, practitioners often question this measure mainly for its invalidity if the distribution of fund returns is beyond normal (Kao, 2002; Amin and Kat, 2003; Gregoriou and Gueyie, 2003, Cavenaile, et al, 2011, Di Cesare, et al, 2014). Several new measures have been proposed and investigated to overcome this limitation of the Sharpe ratio, however, Eling (2008)

finds choosing a performance measure is not critical to mutual fund evaluation, Eling and Schuhmacher (2007) compare the Sharpe ratio with 12 other measures for hedge funds and conclude that the Sharpe ratio and other measures generate virtually identical rank ordering, despite the significant deviations from normal distribution. Similar evaluation includes Eling and Faust (2010) on funds in emerging markets, Auer and Schuhmacher (2013) on hedge funds, and Auer (2015) on commodity investments.

This paper proves that several widely used performance measures are monotonic if the distribution of asset returns is a LS family, a family of univariate probability distributions parametrized by a location and a non-negative scale parameters that is commonly applied in finance (Levy and Duchin, 2004). Our proof certifies the empirical findings in other studies on the indifference of choosing a performance measure when valuing a fund. We show that those measures generate virtually the same rank ordering using monthly mutual fund return data from 1997 to 2005 and Monte-Carlo simulations. Therefore this paper contributes to both the academia and industry by clarifying the phenomenon.

For example, the below figure plots the correlation and confidence intervals based on 2000 simulations for each sample size. For simplicity, we show the results for the Sharpe (ρ1), the Sharpe-Omega (ρ2) and the Sortino ratio (ρ3) only. Consistent with the previous finding, the rank correlation among these performance measures is roughly equal, and is approaching one with the increase of sample size.

Dec
7

Pawel wrote a great article on predicting heavy and extreme losses in real-time for portfolio holders, the goal is to calculate the probability of a very rare event (e.g. a heavy and/or extreme loss) in the trading market (e.g. of a stock plummeting 5% or much more) in a specified time-horizon (e.g. on the next day, in one week, in one month, etc.). The probability. Not the certainty of that event.

Read this excellent post and accompanying Pathon codes at http://www.quantatrisk.com/2015/06/14/predicting-heavy-extreme-losses-portfolio-1/

In this Part 1, first, we look at the tail of an asset return distribution and compress our knowledge on Value-at-Risk (VaR) to extract the essence required to understand why VaR-stuff is not the best card in our deck. Next, we move to a classical Bayes’ theorem which helps us to derive a conditional probability of a rare event given… yep, another event that (hypothetically) will take place. Eventually, in Part 2, we will hit the bull between its eyes with an advanced concept taken from the Bayesian approach to statistics and map, in real-time, for any return-series its loss probabilities. Again, the probabilities, not certainties.

Read this excellent post and accompanying Pathon codes at http://www.quantatrisk.com/2015/06/14/predicting-heavy-extreme-losses-portfolio-1/

Nov
13

I have written a working paper on CDS (credit default swap) implied stock volatility and found some interesting results. Post it here just in case someone is interested.

Both CDS and out-of-money put option can protect investors against downside risk, so they are related while not being mutually replaceable. This study provides a straightforward linkage between corporate CDS and equity option by inferring stock volatility from CDS spread and, thus, enables a direct analogy with the implied volatility from option price. I find CDS inferred volatility (CIV) and option implied volatility (OIV) are complementary, both containing some information that is not captured by the other. CIV dominates OIV in forecasting stock future realized volatility. Moreover, a trading strategy based on the CIV-OIV mean reverting spreads generates significant risk-adjusted return. These findings complement existing empirical evidence on cross-market analysis.