Fuzzycrawler

Fuzzy Logic Search Engine

Structured data that is labeled is the traditional input for modeling anything from economics to biology. With such unambiguous data, you can examine and test relationships between those variables using statistics.

What can you do with data that isn't nicely labeled or even structured? You don't really know what that data refers to or what it means. Is that data effectively useless? That is the central challenge of trying to classify information that isn't already labeled. A key ingredient to such efforts is side-channel data. This is data associated with a document like an encrypted message or an HTML webpage that can be leveraged to help you discover what that document might be about.

Of course, such inferences are inherently uncertain but still valuable. Classical logic can't do much with such uncertainty but fuzzy logic can make a lot of hay with such inputs. The neural networks that power most artificial intelligence today are a way of accurately modeling complex scenarios by differentially and even dynamically weighting such uncertain fuzzy inputs.

I spent years during and after college developing a fuzzy logic based classification engine that crawled and classified documents like web pages and USENET messages. The classification engine used neural networks to derive a consensus classification of arbitrary documents. When I eventually released it as a search engine, its usage exploded, and I struggled mightily to keep it from crashing. As a result, I soon sold Fuzzycrawler.

Ever since I have had mixed feelings about that. I wish I'd known then what I know now about valuation, investments and entrepreneurship. I probably would have tried more to manage and grow Fuzzycrawler. One of my current goals is to teach others about valuation, investment and entrepreneurship to provide them with the evaluation and decision tools I wish I had when I was young. Nevertheless, Fuzzycrawler provided me with a great foundation going forward and a lot of positive reinforcement to keep innovating.              

Economics Natural Experiment

How do you find needles in census haystacks?

Assigning individuals to groups and then tracking changes to them across time is the gold standard of experimental design. To accomplish this, each individual in the experiment is uniquely identifiable by label, number or code across the entire duration of the experiment.

Can you still accomplish individual tracking across time if individuals have not been assigned a unique identifier? Yes, by using those same techniques developed and deployed with Fuzzycrawler. For my dissertation I was able to use neural nets to identify and then track tens of thousands of individuals from 19th century Wales as they traveled across decades of time and thousands of miles to America. And for many of them, back again to Wales.

Each cohort individual in Wales was identified using a combination of government records like censuses and church records like membership rolls. Then, using neural nets, most of those individuals were able to be subsequently identified in American records like censuses and church records. Further, I had additional data on these individuals like occupation, wealth, family and geography and could thus control for these potential confounders.

For this research, these individuals had self-assigned to one of two groups: those who migrated to cosmopolitan industrial Pittsburgh and those who migrated to the remote swampland of Venedocia, Ohio. This tracking and classification allowed me to examine a natural experiment of the social effects of British economist Ronald Coase's famous Coase Theorem about transaction costs and the social cost of externalities. My results confirmed Coase's theoretical insights and contradicted existing modeling of migration flows that presume economic migration volume is tied to international business cycles and is typically one-way.

Both locations burdened new immigrants with significant environmental externalities. In Pittsburgh, there were significant transaction costs to organizing that prevented a solution to that social cost. Consequently, the overwhelming majority of Welsh migrants to Pittsburgh returned to Wales. In Venedocia, however, there were not significant transaction costs associated with self-organizing a social response to that social cost. So, a solution was found and those immigrants stayed permanently. Transaction costs shape society in profound ways.            

Lehjr

Financial Statement Fundamental Analysis

Maybe like me you've noticed the considerable gap in quality between free financial information and ratios and the financial information available for thousands of dollars per month? Why does something as foundational as Beta differ between Yahoo Finance and Google?

Do I really need to spend almost $30,000 a year to get reliable and quality financial information? Most of us don't need the firehose of information provided by a Bloomberg terminal. Nor can most of us afford to rely on the accuracy of low-end financial data providers.

Lehjr aims to fill that gap. Lehjr is positioned to be the value maximizing financial analysis platform. 

By sourcing data directly from SEC filings and maintaining a laser sharp focus on data quality and fundamental financial statement analysis, while foregoing frivolous add-ons, Lehjr will provide investors with a Toyota-like value proposition. Lehjr will be the unfailingly reliable and eminently sensible high value platform for individual investors available at a reasonable cost.   

Lehjr is currently in Beta testing. 200 people are now road-testing every aspect of the platform, which is expected to launch in Q4 2025. Check it out at Lehjr.com

Bayesian Global VAR

How do you improve macroeconomic forecasts?

Since their inception nearly 100 years ago, macroeconomic models and forecasts using econometrics have been panned as hit and miss. They were hailed during the 1920s but were all but forgotten during the Great Depression. They were reignited during the Keynesian revolution and regained prominence during the 1960s only to fall out of favor again in the 1970s for their failure to forecast stagflation. 

Since then, numerous modifications have been made to limited success, but all such macroeconomic modeling suffers from the same causal modeling limitations. They work well enough when conditions are well understood and specified, but they tend to be utterly flat-footed when those conditions take unexpected turns.

Data science offers economics a paradigm shift in forecasting. Machine learning trained on economic data is consistently able to outperform conventional econometric approaches.

I'm developing a Global Vector AutoRegressive model of global trade flows that forecasts macroeconomic variables of interest using a Bayesian statistical approach. Early results are very promising. 

Small Church

How valuable can expert financial advice be for an organization?

A small church had just reached a legal settlement to leave their denomination with their church building by agreeing to pay them $100,000 every year for the next 10 years. If they ever missed a payment, the building would revert back to the big denomination.

The church had fundraised the entire $1 million prior to the settlement. Its board wanted to payoff their exit fee immediately to avoid the risk of future nonpayment. They were worried, frustrated and felt out of their depth.

So I advised them to purchase a 10-year annuity that pays $100,000 every year. The cost of that annuity was only $930,000, and the nonpayment risk was now covered.

By saving $70,000, my advice was theoretically worth that much to them. But a few years later that same small church had a significant cashflow problem and that $100,000 annual payment would not have been paid. Thanks to my annuity advice solving their risk problem, their building was saved. The real life value of expert finance advice can be the organization's continuing viability. My personal takeaways are that stochastic processes are real alongside a reaffirmation of my father's advice that a failure to prepare is a preparation for failure. 

U.S. Coast Guard

How do you value a volunteer organization?

The US Coast Guard Auxiliary is an all volunteer uniformed part of the US Coast Guard formed by Congress during WWII. There are more than 30,000 Auxiliarists spread across all 50 states.

The only Coast Guard tasks that an Auxiliarist is prohibited by law from performing are law enforcement and combat ones. So these Coast Guard volunteers perform an incredibly wide array of jobs from flying their personal planes to vessel inspections to research and development.

For years Coast Guard leadership wanted a valuation of their Auxiliary's diverse contributions of manpower and equipment, but they hadn't been able to perform this valuation in a way that passed DHS or Congressional scrutiny. I took over this project and created a valuation methodology that passed all Washington DC scrutiny so well I literally received a big medal for it.

The Auxiliary valuations I performed made their way into the Commandant's speeches and the first significant budgetary increases for the Coast Guard from Congress in years. The methodology I created is classified and still in use by the Coast Guard to this day.

From my award pinning ceremony: "You applied remarkable expertise in quantitative research to formulate a new statistical methodology for analyzing Auxiliary activity data. Exercising exceptional insight, you ... helped keep the highest levels of Coast Guard leadership clearly appraised of the extraordinary value of the Auxiliary's uniformed volunteers."        

Economics at Work