Applying statistics and machine learning to solve complex business challenges.
Our primary mission is using data to discover and exploit opportunities in a wide range of scenarios.
We are data scientists, business experts, financial modelers, and general information ninjas. The main areas of our focus are artificial intelligence, financial modeling and computational advertising.
With over 12 years of experience developing machine learning systems, we are intimately familiar with a wide variety of methods.
From sophisticated multi-strategy back testing, to advanced algorithmic trading, we can develop financial modeling and execution systems.
We have a developed proprietary algorithms and systems to optimize online advertising for maximal ROI.
The core of our practice comes from a deep understanding of statistics and statistical theory. We are both Bayesian and Frequentest as necessary.
Successful strategies are only as good as their implementation. We are well versed in a variety of advanced computational techniques, including C++, R, Perl, Matlab, CUDA, Parallel Algorithms, and more.
Data is the key to your world. We understand large data, high speed databases, redundant storage, and other idiosyncrasies of data management.
A common thread amongst all our practice areas is the ability to accurately classify things based on their features. We have a arsenal of maximal margin classifiers to fit almost any need.
Boosted algorithms are a very successful technique that perform well in a large variety of applications. We are well versed in boosting techniques, including ADABOOST and REALBOOST.
Analyzing documents for classification, concept extraction or automatic summarization can be challenging. We have experience with TFIDF, mutual information criterion, summarization, feature reduction, and other advanced statistical methods.
With experience in both medical research, marketing, and finance, we are well versed in developing properly designed experiments to minimize bias while studying the question.
We have experience designing algorithmic trading systems using a variety of technical indicators and broker APIs.
We find that hierarchical Bayesian models fit many problems well. We can execute sophisticated Bayesian models using either a Gibbs sampler or Markov Chain Monte Carlo.
One significant advantage of online advertising is the incredible level of detailed data available. Unlike TV, radio, or print, we can see ad impressions and responses on an individual level, in real time. We can track who sees an ad, if they visit your website immediately, or at any point in the future. We can correlate impressions and/or repeated impressions by demographic with website visits or other goals. However, most brands are still using very simple static ad-hoc models that fail to leverage all of the information available.
Ads placed within Google search results are a significant strategy for many advertisers. However, the optimization of those ads and budgets can be extremely complex.
We have developed multi-armed-bandit based algorithms that borrow from modern portfolio theory. Our algorithms automatically re-balance your ad budgets and creative mix to maximize return on spend.
The most exciting development in online advertising is the growth of Real Time Bidding. The Smart Media Corporation is an approved bidder with a seat on the Doubleclick Advertising Exchange.
We are currently developing new layers of artificial intelligence to help brands optimize their online advertising effort. Our proprietary algorithms and systems allow us to optimally bid on individual ad displays per person on target websites. This is an entirely new world of behavioral based advertising that provides powerful ways to maximize return on advertising dollars.
A common challenge in advertising is the question of how to attribute ad spend to brand image or consumer action. Consumers are often exposed to a mix of many online and offline ads. Determining both the individual and collective impact of those ads is very difficult.
We have developed sophisticated hierarchical models that allow us to measure the offline impact of online advertising and predict consumer behavior as a result of the viewed advertising mix.
The rapid growth of technology has made computational finance and algorithmic trading more accessible to both smaller firms and individual investors. There are a wide range of tools available, from simple Excel spreadsheets, to full trading systems provided by online brokers. While the off-the-shelf tools are fine for the amateur investor, most need more powerful and flexible solutions.
We can help you develop complex strategy back-testing systems, portfolio optimization, full algorithmic trading systems, or anything in between. Our team generally works with R, Interactive Brokers, or raw C++ as needed. However, we are open to developing financial models and systems within your preferred platform.
Modern Portfolio Theory, introduced by Markowitz in the 1980′s has been a standard in the financial industry for decades. Recently, research has moved beyond MPT to better ways of maximizing profit while reducing risk. We have worked with Monte Carlo based portfolio optimization, and an optimal Kelly fractional portfolios, both of which tend to perform better than MPT. Another key issue is that of asset variance, which all portfolio optimization algorithms need. We’ve innovated new dynamic estimating tools that use discrete time sliding windows to better measure variance.
Algorithmic trading involves the use real-time computation to estimate the direction of an asset, a pair of assets, a derivative, or an index. We have developed numerous systems to implement client strategies while connected directly to a broker’s data and trading API. Some systems have been interesting combinations of traditional technical indicators, while other systems have involved our own dynamically calculated derivatives and measures of market or asset factors.
Calls and puts are standard options most are familiar with. There is an entire universe of futures and derivatives far beyond these two. We have worked with futures, options on futures, index options, index futures, commodity futures contracts, and even energy derivative trading. Pricing both simple and exotic derivatives is usually done via Markov Chain Monte Carlo methods, a tool we have much experience with. Our advance parallel computing platform allows for very rapid future price calculation, enabling trade decisions to made quickly.
We would love to discuss your project specifics and how we can help. Please contact us to start a conversation.
8149 Santa Monica Blve. #176
Los Angeles, CA 90046
(323) 653-1900
support@smartmediacorp.com
