Please tell me about Lucena Research.
My name is Erez Katz. I am the CEO / Co-Founder of Lucena Research. I have teamed up with Tucker Balch Ph.D, who was at the time a professor at the Georgia Institute of Technology. Over time we’ve grown to become experts in the space of machine learning for the financial markets from both academic and business sense.
Lucena’s Quant Desk. Source: Lucena Research
Since inception in 2013, we have been working to connect big data and alternative data providers with investment professionals who are seeking to extract actionable information from data to make better investment decisions and for KPI forecasting. Our company started by creating a technology platform. called Quantdesk. The toolset harnesses machine learning capabilities and we make them accessible to the non-quant and the non data-scientist. Professional investors who are looking to deploy data intelligence can use our platform to extract information from data all with a click of a button. They don’t have to be tech savvy to access our technology. We help augment the decision-making process using machine learning and big data science. That’s our mission.
We have developed a breath of offerings and services that is scaled to support young, emerging hedge funds and family offices, all the way up to the most successful and sophisticated hedge funds and financial institutions.
What were some of the challenges you faced?
Credibility was the biggest challenge for us. Firstly, when we first started, machine learning and data science was not in the forefront of the media and not in the objectives of some of the bigger companies we were trying to engage with.
Secondly, when you’re an emerging company in a financial market space, pedigree and history as part of your experience or working as a professional investor in the field, goes a long way. In the early days, our company was made up of mainly technologists who came from academia and/or with business acumen. But not necessarily with a financial background. That was a challenge for us because the financial landscape found it difficult to view us as a reliable source.
But this is one of the things where perseverance and consistency goes a long way. It’s just a function of time. We kept on educating the market through networking, presenting and providing thought leadership. Overtime, we became experts in the space not just in machine learning, but more importantly, how to apply machine learning in the context of KPI forecasting and price forecasting for assets.
Lucena’s Backtest Performance Report. Source: Lucena Research
One of the things that we had to do was to provide empirical evidence that our data science worked. On our platform, we have a comprehensive suite of reporting systems, backtesting systems and perpetual paper trading simulation systems, that allows our customers to view our forecasting decisions in real time and evaluate the efficacy of our technology. In essence, we gave users the ability to evaluate how a forecasting for a given asset universe has turned out after we’ve made a projection. Our transparency into what works and what doesn’t, created additional interest and built that needed trust in not only our technology, but also our outcomes.
It’s a staying power game. You need to be strong enough, live long enough to have the market turn your way. In the last 2 years or so, we see a massive shift in adoption of big data and machine learning technology. It’s all in its infancy but the market is coming to us which is a very exciting time for us.
Any tough pitches/meetings you recall?
One day, I had a meeting with a professional investor from a very large consumer based platform. It cost me a 2 hour flight and a stay at a hotel. When I finally met him, he said to me you have 10 minutes. That was the first thing he said. Didn’t say hello, didn’t shake my hand.
So the people that we work with are very busy. It’s a very stressful job. Having to listen to another story that they’re not really comfortable with, was a big challenge even for them because they have to balance their time with their curiosity. But now, the market has matured and has become more accepting of our offerings, the situation is vastly different. People are calling us asking for our time to have these discussions and it’s grown a lot more favourable for our business.
Any point in your entrepreneurial journey where you had to make a tough call?
We have to make hard decisions everyday. As an emerging company in an emerging space, there is a point where you have to say no to new business that you don’t think is going to enhance your long term plan. When you are a company of our size, we did not raise a lot of capital, we are growing organically where every income means cash flow. We had an opportunity in the past to engage with someone who was to provide a very strong income stream for our business but it was to work for a company that was selling mortgage services. We decided to say no which was a very hard decision for us to make because it was income that we were essentially giving away. But we knew it was not the right strategic move for us at the time. In hindsight, we are glad we made that decision because we remained true and strong to our core competency and not trying to be everything to everyone. I think it’s very important for every business owner to learn when to say no to opportunities that provide immediate cash flow but not a long term perspective for the business.
How do you differentiate from your competitors?
One of the differentiators is our transparency. Frankly, there is a lot of misinformation in the space. People use the term AI, Machine Learning, Big Data very loosely without actually knowing what it means.
One of the other things we pride ourselves in is our technological capability. We don’t just talk about what we can do for them, we actually have a strong platform where users can sign up and very quickly identify all of its capabilities intuitively. It’s very easy to use which I believe sets us apart. Many other providers tend to rely on professional services which is much more expensive and the outcome may not necessarily be as clear. But with us, we draw clear outcomes.
More importantly, we are very customer centric. We listen to what our customer needs and we can quickly augment our offerings to specifically support what they’re looking for as opposed to what we have to sell them. That’s another big advantage of our company.
Lucena’s Data Matching Engine. Source: Lucena Research
Let me give you an example. We have very large datasets within our database. It’s hard for users to interpret which datasets are going to be reliable, and if it is reliable, how can I use it in the context of my specific needs. So we have built a module on our platform called “Data Matching Engine”. It takes you through a set of screening questions, where it asks you to upload the constituents universe that you’re most interested in, about your investment style, what your type of portfolio is, your time horizon of investment and a series of other questions. This takes place through a ‘wizard’. Once you push the submit button, the system goes out and identifies from all our datasets, all our factors, the data that would be most suitable for the specific scenario you just outlined. The results will showcase the datasets that are most appropriate for your specific needs and how these datasets can be used in the following context. We also show the user empirically why did we decided that these datasets are going to be more suitable for their parameters.
What datasets are you working with?
We have quite a few partners. Some large data providers we work with are IBM, ADP which is a payroll company, Equifax which is a consumer credit company. But we have a plethora of small data providers that are very unique that provide data on social media sentiment, corporate earnings findings, insider buying and selling, social media consumer emotion gauge that identifies how emotionally engaged people are in a subject matter. The list goes on.
By having all these datasets together, it presents us with a very compelling opportunity to create a multifactor model, meaning that once you aggregate the pieces of information from different datasets that can coexist together, you can be provided with a much higher conviction when forecasting, than by just an individual dataset by itself. Our technology is unique in the context of identifying these relationships between octagonal datasets very effectively.
Lucena’s Alternative Data. Source: Lucena Research
Think about regular fundamental data that are mainly driven from corporate balance sheets and earnings report. You take that and combine it with technical data like price movements, mean-reversion and momentum scenarios. Then you further combine it with macro economic data which is all part of our baseline dataset. Now you take all that and add a whole variety of alternative data companies that have unique offerings of their own.
Lucena’s Data Validation Process. Source: Lucena Research
From here, we don’t analyze just any data. We have a validation process that allows us to screen which data is really worthy of the user’s consideration for further research. Some of the datasets are either not mature enough, they lack history, or the data is not ‘survivor bias free’, which is basically how do you consider historically stocks that no longer exist in your data. There are many ways that we test the data to make sure it is authentic, reliable and complete. It is a big problem to solve and we’ve streamlined many of these processes through automation.
Who are your clients?
In one respect, we have many data providers as clients but on the buy side I would say there are three tiers. The top tier are the quant funds. These are the most sophisticated quantitative analysis teams whose job is to identify alternative data and how to mobilize it for their internal needs. These companies are not necessarily going to benefit from our quantitative research capabilities because they have their own. But they do consume our data validation services. If they have a 100 datasets in their queue, how would they know which one should be considered first? They essentially want to fail fast. They don’t want to spend 3 months into quantitative research to realise that what they were going after was the wrong dataset. We give them a higher degree of success through our validation process. That’s the top tier.
Lucena’s Data Review Engine. Source: Lucena Research
The second tier is the deep value fundamental funds that are not quantitative naturally. But they’re trying to dip their toes in the water. They’re penetrating slowly and carefully into the quantitative space. We call them quantamental. They’re fundamental in nature of their business but they’re trying to augment their business by introducing quantitative measures. These clients use us in a more involved way for KPI forecasting or smart data feeds, because they want the signal. They need us to make sense out of the raw data. So we interpret the data for them and give them a more actionable signal out of the data.
Lucena’s Model Portfolios. Source: Lucena Research
The third layer are the emerging funds. These are hedge funds, family offices and other investment professionals that are trying to create portfolios that are completely algorithmic. We provide them with model portfolios that generate trading signals, trade simulations of a portfolio, to help them take advantage of the data, the science and our research for a complete algorithmic trading strategy.
Which of your products are most popular?
Our platform is very robust, QuatDesk(r) and there is a lot of demand for the “do-it-yourself” researcher. Recently however, we see great push for KPI forecasting. Not forecasting stock prices of assets or macroeconomic trends in the market, but key performance indicators for big companies. It could be anything from change in gross sales to profit margins all the way down to return per unit. It also works for airline ticket pricing projections, total sales of consumer goods and real estate forecasting or pricing of homes
Lucena’s KPI Forecasting. Source: Lucena Research
Our ability to forecast KPIs has a much higher degree of accuracy than price forecasting. The reason is when it comes to asset prices, the data is very noisy. If you get a 55-58% accuracy, you’re doing fairly well. When you forecast KPIs, you can easily get up to 85-95% accuracy, which is a much higher degree of reliability. On top of that, many of our customers have their own secret sauce or their own way of doing things. Once they receive the KPIs, which form the underlying ingredients of their menu, they can then go off and cook the meal for themselves. Meaning they can actually forecast the actual portfolio themselves if they were given the underlying KPIs or the underlying ingredients, to make that final decision.
Let’s say somebody wants to identify what’s the projected real estate pricing for homes in certain regions in the US or in the world for that matter. Once they come onto our platform, they can upload their historical home prices by quarter or by month over the last few years. So users give us a time series that represents the home pricing over time. In turn what we do, we identify which factors from our list of 950 factors from multiple datasets are most applicable for what they have uploaded to us. We then use our system, it’s called “machine learning classifier”, to classify which factors together have yielded the highest probability of historical KPI forecast, and use these factors to forecast the next quarter complete with our reasons for the forecast.
Lucena’s Data Qualification Engine. Source: Lucena Research
The opportunity transcends a much broader market than the financial market. Think about any company that wants to project their future KPIs. It could be retail companies projecting growth in sales based on alternative data that belongs to them. For example, credit card receipts, consumer footprint in a store, shelf life and the turnover inventory. All these datasets can be aggregated very much like how we aggregate alternative data for investment, but this way, we are adding data to allow companies to make more informed decisions about their future using AI and machine learning technology.
Where is machine learning at today and what challenges does it face?
We pride ourselves as a leader in machine learning technology. We cover a vast spectrum of machine learning capabilities, from traditional modeling such as KNN, decisions trees, logistic regression, and SVM, all the way to deep learning technology that uses convolutional neural network and long and short term memory, LSTM with RNN. We have a pretty comprehensive expertise on all assets. I think it’s a function of data. The most sophisticated machine learning capabilities like computer vision, self driving cars or facial recognition, all these key technologies need to have an abundance of sample data or labelled data. This is how you train the machine to generalize features that can be used in the context of future data that has yet to be seen.
With machine learning applicability for KPI forecasting, think about KPI of net sales. Data is very scarce. You only have data points once every quarter you need to have millions of sample data, for machine learning to effectively learn how to trade stocks. In many cases you have to use a simple model because the data is not as abundant as other spaces or other sectors of the economy. Even as the amount of sample data increases, the technology we currently have is already ahead of its time in the context of machine learning and deep learning and reinforcement learning.
With alternative data, some of it weren’t available before. Social media has only started to become widely used in late 2013, so you only have 4 to 5 years of data. That’s not enough to create deep learning capabilities but in the context of simple machine learning models, they’re quite effective.
What do you think of Asian data?
We are very envious of the Chinese AI market because there is a very broad universe of data that is available for machine learning research. There is a challenge in the US and Europe about data privacy that can create some obstacles from pure research perspectives. I know that in China and some Asian countries that problem is not as restrictive as in the US and that provides a very big advantage for the Chinese market in Machine Learning and Big Data. We definitely see China as a very strong competitor to the US market. But ultimately, this technology is really transformative. If you think about the big data and ML, it’s as big as many other industrial revolutions in the past. I think it is a world problem we’re trying to solve, in order to make the world a better place. I think China is a very strong competitor but also a very strong collaborator in trying to achieve breakthroughs in the aspects of research, big data and machine learning.
Do you foresee Lucena to become active fund management leveraging the technology that you have?
That’s one of the most common questions people ask us. Mostly they ask because they are skeptical. It’s really important to make a distinction. I mentioned before the degrees of accuracy. When you have 58 to 60% degrees of accuracy when you forecast stocks, you’re not going to be right all the time. There’s a lot more than just getting a predictive signal to run a successful hedge fund or successful money management.
Currently, many of our customers are hedge funds that are looking for us to engage on a partnership with them on particular investment vehicles that have the upside potential for us, if we were to be successful. We’re also engaged with several companies on these same grounds. But ultimately, our company wants to stay a tech company. We want to be pioneering new frontiers, new research ideas, new capabilities, as this world of ML and big data continues to evolve at such a rapid pace. This is our passion and this is what we love to do.
Other people love to invest, to use investment vehicles, to deal with customers and compliance and the ups and downs of the market. We decided to stay away from that and become a pure technology play because this is what we know how to do. But to answer your question, we are, figuratively speaking, eating our own food because we are engaging companies that are investing using our technology. We have what we call an upside reward if we were to be successful in how we drive these strategies. So indirectly, we do that. We are investing using our own money even though we’re not the actual people behind it pushing the buy-sell button. It’s our customers who do that.
What’s in store for Lucena?
We are as a company going through an inflection point. We’re growing and we have to decide how we’re going to capitalize on the opportunity the market is giving us.
Lucena’s Partnership Strategy. Source: Lucena Research
We have a few opportunities now. Number 1 is to raise capital and become a much larger company. Number 2, a strategic partner to work together to expand rapidly. Number 3 is to get acquired by a large outfit that is looking for our capabilities as a strategic investment for their outfit. So we are evaluating all these three options. One is to grow organically. One is to merge and the other is to get acquired. We have active conversations on all of these fronts. We would love to figure out a way to partner with somebody who has access to the Asian market.
Lucena Research website: https://lucenaresearch.com
Erez is the CEO and Co-Founder of Lucena Research. Erez was instrumental in architecting Lucena’s flagship platform QuantDesk®. Prior to Lucena, Erez was the founder of Objectware Inc, an enterprise cloud software company, which was sold in 2007 to Bridgeline Digital. Erez assumed Bridgeline’s Chief Operating Officer through 2011 and was instrumental in its growth and ultimately IPO on the Nasdaq stock exchange.