I received a book I ordered from Amazon today titled : "Neural Networks In Finance" by Paul D. McNelis.
It concerns the automated sifting of our personal data in relation to credit risk and forecasting our financial future.
In a nutshell, the author and his team (all statisticians and financial research scientists) tested current computer technology by using neural networks (software) to see if computers can better predict our finances based on various criteria.
The answer is a resounding yes. And the computers do it faster and more accurately than previous systems.
The programming is heavy on statistical analysis and mathematical theorems, an area where I have no training. But the conclusions are spelled out in fairly easy to read English.
These systems are currently in use by major banks, credit card companies, Fair Isaac Corp. etc. The next generation of this will be life-altering, I suspect.
They dump every possible bit of info into massive computers and use them to build comparisons. There ARE in all probability, no matter how unique your life, many people who are a statistical "twin" to you and your financial choices. This methodology allows banks, card companies, financiers, anyone who might extend you credit, the ability to forecast with a 98% immediate success rate your chances of defaulting, nonpayment, or bankruptcy. The success rate at predicting years down the road drops somewhat, but only if they do not update the information. Unfortunately it is extremely easy for them to update. In fact it is automated. If you move, for example, or change jobs, they can generally see that information almost immediately, in one fashion or another.
The information thrown into these programs is staggering.
Some of the criteria culled from public records and databases, as well as websites and your purchases include:
marital status
home ownership
income (and income history and future forecast of your income)
gender and race (thought this was illegal)
address
number of children
employment (done by code numbers for virtually every job imaginable. Even mine, as self-employed performer)
medical history (including factors like weight, height and obesity, insurance company information, past health issues and current medical records)
your address
recent moves in residences or employment
your phone number and area code-as a comparison to others
FICO
number of people in your home (dependents and non-dependents)
whether the phone is in your name
shopping patterns (ie-takes a vacation every summer, but didn't this summer. RED FLAG)
Average cost of home/rent in your neighborhood
Neighbors occupations
Rolling average of your real estate values and rate of return on the investment in home or land
Receive mail at home or at PO box
Age and genders of children
Rent or own vehicle (what type and age as well as purchase history)
distance to work
hours worked per week
citizen or foreigner
There are OVER 1,000 criteria like these used for every single one of us.
Some models rely on "crisp logic" and others use "fuzzy logic" as defined broadly:
Crisp logic categorizes you exclusively into one subset and no others. For instance: you are overweight by 8 lbs but still within the recommended weight range. The computer would define you as either obese or not obese.
Fuzzy logic: You can be categorized multiple ways. In this example, you would be listed as both overweight AND normal weight, with the 8 lbs difference applied as a percentage to each category-perhaps 40% of the score into overweight and 60% into the normal weight category.
I know this is tough reading, but bear with me. We are nearly to the really important stuff.
Think back through your life, to all the times you signed "disclosure agreements". With a school, medical provider, job, whatever. All that information is STILL there, and it is rapidly being scanned and fed-at least as much as they can get. Without a doubt, they certainly can't have a complete picture of all of us. But they have enough. The difference between the computer models having 300 answers instead of a thousand is a loss of below 2% in their accuracy.
Here's the important stuff:
As we move forward, and computers become ever faster, ever more powerful:
Companies will be requesting even MORE information from us.
New records will be easily available to automated software that is designed to scan through OTHER databanks that store everything from financing records, credit card statements, school records, employment, etc. The records of the past might be unavailable, since there weren't computers to scan them into a few years ago, or the manpower necessary to begin with.
Not the case these days. Almost everything we do now ORIGINATES in computer formats.
It means that these companies, as we move into the future, will possess an incredible amount of power over us, our finances and children.
Also, the computer programs themselves are designed to "learn" from their mistakes. Let's say you are an anomaly. You are making a million bucks a year and default on a 15k loan. The neural network will note it originally, as a failed prediction. Then it will file that info away until/unless it happens again with someone matching your specific criteria. Based on THAT, it will make comparisons to others who are similar and may be able to locate intertwining, hidden circumstances that allow it to predict the same scenario before the THIRD person defaults in a similar situation.
The computers are constantly correcting their own errors, getting "smarter" as time goes by and more information is available. This does not mean (Ihope) that the computers will "take over the world" or some other conspiracy gobbledy-gook. It means the software programs are simply becoming more effective at figuring out what we will do BEFORE we do it.
The programs have many other functions ranging from predicting churn in stock markets, forecasting inflation, factory orders, and general economic work. I am concentrating on the debt aspect, since that is what we are all here for.
I will update this as I read more.
DMC
It concerns the automated sifting of our personal data in relation to credit risk and forecasting our financial future.
In a nutshell, the author and his team (all statisticians and financial research scientists) tested current computer technology by using neural networks (software) to see if computers can better predict our finances based on various criteria.
The answer is a resounding yes. And the computers do it faster and more accurately than previous systems.
The programming is heavy on statistical analysis and mathematical theorems, an area where I have no training. But the conclusions are spelled out in fairly easy to read English.
These systems are currently in use by major banks, credit card companies, Fair Isaac Corp. etc. The next generation of this will be life-altering, I suspect.
They dump every possible bit of info into massive computers and use them to build comparisons. There ARE in all probability, no matter how unique your life, many people who are a statistical "twin" to you and your financial choices. This methodology allows banks, card companies, financiers, anyone who might extend you credit, the ability to forecast with a 98% immediate success rate your chances of defaulting, nonpayment, or bankruptcy. The success rate at predicting years down the road drops somewhat, but only if they do not update the information. Unfortunately it is extremely easy for them to update. In fact it is automated. If you move, for example, or change jobs, they can generally see that information almost immediately, in one fashion or another.
The information thrown into these programs is staggering.
Some of the criteria culled from public records and databases, as well as websites and your purchases include:
marital status
home ownership
income (and income history and future forecast of your income)
gender and race (thought this was illegal)
address
number of children
employment (done by code numbers for virtually every job imaginable. Even mine, as self-employed performer)
medical history (including factors like weight, height and obesity, insurance company information, past health issues and current medical records)
your address
recent moves in residences or employment
your phone number and area code-as a comparison to others
FICO
number of people in your home (dependents and non-dependents)
whether the phone is in your name
shopping patterns (ie-takes a vacation every summer, but didn't this summer. RED FLAG)
Average cost of home/rent in your neighborhood
Neighbors occupations
Rolling average of your real estate values and rate of return on the investment in home or land
Receive mail at home or at PO box
Age and genders of children
Rent or own vehicle (what type and age as well as purchase history)
distance to work
hours worked per week
citizen or foreigner
There are OVER 1,000 criteria like these used for every single one of us.
Some models rely on "crisp logic" and others use "fuzzy logic" as defined broadly:
Crisp logic categorizes you exclusively into one subset and no others. For instance: you are overweight by 8 lbs but still within the recommended weight range. The computer would define you as either obese or not obese.
Fuzzy logic: You can be categorized multiple ways. In this example, you would be listed as both overweight AND normal weight, with the 8 lbs difference applied as a percentage to each category-perhaps 40% of the score into overweight and 60% into the normal weight category.
I know this is tough reading, but bear with me. We are nearly to the really important stuff.
Think back through your life, to all the times you signed "disclosure agreements". With a school, medical provider, job, whatever. All that information is STILL there, and it is rapidly being scanned and fed-at least as much as they can get. Without a doubt, they certainly can't have a complete picture of all of us. But they have enough. The difference between the computer models having 300 answers instead of a thousand is a loss of below 2% in their accuracy.
Here's the important stuff:
As we move forward, and computers become ever faster, ever more powerful:
Companies will be requesting even MORE information from us.
New records will be easily available to automated software that is designed to scan through OTHER databanks that store everything from financing records, credit card statements, school records, employment, etc. The records of the past might be unavailable, since there weren't computers to scan them into a few years ago, or the manpower necessary to begin with.
Not the case these days. Almost everything we do now ORIGINATES in computer formats.
It means that these companies, as we move into the future, will possess an incredible amount of power over us, our finances and children.
Also, the computer programs themselves are designed to "learn" from their mistakes. Let's say you are an anomaly. You are making a million bucks a year and default on a 15k loan. The neural network will note it originally, as a failed prediction. Then it will file that info away until/unless it happens again with someone matching your specific criteria. Based on THAT, it will make comparisons to others who are similar and may be able to locate intertwining, hidden circumstances that allow it to predict the same scenario before the THIRD person defaults in a similar situation.
The computers are constantly correcting their own errors, getting "smarter" as time goes by and more information is available. This does not mean (Ihope) that the computers will "take over the world" or some other conspiracy gobbledy-gook. It means the software programs are simply becoming more effective at figuring out what we will do BEFORE we do it.
The programs have many other functions ranging from predicting churn in stock markets, forecasting inflation, factory orders, and general economic work. I am concentrating on the debt aspect, since that is what we are all here for.
I will update this as I read more.
DMC
Comment