Editor's note: This is Part 3 of a special series. Here are Part 1 and Part 2

NEW YORK ( TheStreet) -- OK, if you are still with me, thank you. I want to add a little more complexity and talk about the growth of "mezzanine" and "resecuritization."

You may recall our discussion about the basic senior senior/subordinated structure of 85% senior/15% subordinated. Given that we have a ratings agency methodology for determining the risk profile for a desired rating, can we further tranche the 15% piece? The answer is, sure, the methodology is in place.

Maybe if we want a triple-B rating we don't have to be quite as stringent as six times the normal historical loss rate. Maybe we only need to use four times that rate. As long as there is 10% subordination or credit enhancement, we can achieve a triple-B rating. Maybe the requirement for a single-A rating is five times the historical loss rate. So now we have a four-tranche deal: triple-A for the top 85%, single-A for 12.5% to 15%, triple-B for 10%-12.5% and equity for 0-10% of losses.

If you remember the option market analogy, where the senior tranches are being short a put on the portfolio and the equity is being long a call on the portfolio, how should we think of these two mezzanine tranches? That's a very good question.

These pieces are knock-in levered loss pieces. Once the equity gets eaten away, they become the first-loss pieces. Here, a 2.5% loss to the portfolio now becomes a 100% loss to the triple-B piece. All is fine as long as there is some subordination, but once that gets eaten through, these pieces take great pain. Probability of default is determined, but loss given default is huge. These are kind of like a straddle or strangle with respect to loss rates. As long as losses are between "X" and "Y," you are OK. But if they go outside those boundaries, you would have been better off in the seniors or in the true equity. From an options standpoint, these are much tougher to price given their sensitivity.

I can make an analogy here to corporate bonds and stocks as well. At the bottom of the capital structure of a corporation you have common stock, then you have preferred stock, then subordinated debt, then senior debt and lastly senior secured debt. So the tranching process we have been discussing has legitimate parallels.

But the difference with mortgage pools is that if we tweak an assumption, there is a dramatic effect. With a company, the outcome is pretty binary: it either survives or it defaults, so one doesn't tend to think of the leverage embedded in each of these securities. With a mortgage pool, there are many outcomes that can occur, all of which are magnified by the leverage in the tranche.

So here are lessons No. 2 and No. 3 of the crisis. Given the dependence on an assumption set, we must always, always question our assumptions and our sensitivity to those assumptions.

We cannot assume assumptions are truths; we must recognize that they are assumptions. Also given the severity of what can occur in a mezzanine tranche (which again is a knock-in levered loss piece), I don't think any tranche with embedded leverage should be rated triple-A.

As we continued the tranching, we saw subordinated triple-A tranches, and that was obviously a mistake. A triple-A tranche should never be subject to complete obliteration, no matter how remote the possibility.

The fourth lesson for of the crisis is to be cognizant of diversity, or its inverse, correlation. In one of my early articles on TheStreet, I wrote that it wasn't the asset correlation that went to 1 in the crisis; it was the asset financing correlation that went to 1. When the financing rug got pulled, every asset reliant on financing tumbled. Even in late 2007 we were being told by experts that the crisis would largely be contained to the subprime sector. Wrong!

As long as we are tranching things up, why can't we get into the "resecuritization" of assets? A mortgage Collateralized Debt Obligation (CDO) would pool up existing tranches of MBS assets. Before you say, "OK isn't that a little too far?" remember that the very first CDOs were the exit strategy for all the S&L high-yield bond portfolios, and I do believe they serve a legitimate purpose in portioning out the risk/reward of such a portfolio between investors interested in excess return and those interested in stable value.

But if tranching creates option-like exposure, when we do a CDO of other structured finance investments, we are creating option-like structures on option-like structures. The derivative is squared. This makes the sensitivity to assumptions all the more important. And the issue here is that the rating agencies eat their own cooking.

In other words, if a guy is rating a mortgage CDO made up of a bunch of triple-B-rated MBS instruments, that guy believes the MBS instruments are triple-B risk as rated by one of his co-workers. But "Mr. Market" might price some of those triple-Bs at a 6% yield and others at a 12% yield. So the market itself is telling you something about the perceived riskiness of those two assets. This is a flaw of self-referencing that the agencies can have.

But investors are on to that fact -- which is why they will demand to see what exactly is in the portfolio of a CDO, so that they can make their own determination of risk. They will run scenarios where they may shock the 12%-yielders with early/immediate default and then run a different vector of default on the rest of the portfolio. Above all, if they do not like the portfolio or their stress test results, they will not buy.

But the leverage on leverage heightens the sensitivity to assumptions even further. Let's say I buy a triple-A tranche that has 10% subordination and is backed by a portfolio of triple-B MBS tranches. I am reliant on the those underlying tranches indeed being triple-B rated. That is based on a four-time historical loss coverage. But what happens when the rater decides the historical loss rate is incorrect and should be 5% instead of 2.5%? All of a sudden my resecuritization looks woefully inadequate because what I thought were triple-Bs are now double-Bs. What if that number isn't 5% but 10%? All of a sudden I am short a bunch of in-the-money options. And given that the correlation may be higher than I assumed, 10% subordination on my CDO means nothing. Uh-oh.

But this shows that when you are that levered to an assumption set, it really doesn't matter much about asset selection, just as it doesn't matter much where you are sitting in the tub if the hairdryer falls in the water.

So we can kind of see how each step of innovation and each assumption was built on previous assumptions. It was only when our assumptions about the way the world worked proved wrong that we realized how levered we were to those assumptions.

I have one last comment. Whether it be because of increased complexity or the Efficient Market Hypothesis or the Yale strategy of portfolio management, we have seen an increase in specialization in our markets. Specialization removes flexibility. If I hire a team to specialize in managing and evaluating structured products, it becomes much more difficult for me not to invest in those areas. This is like the NASA theory: If I don't spend my budget this year, my budget gets cut.

It is worth noting that in the infamous Abacus deal, the long side consisted of professionals dedicated to investing in this stuff. The short side was a novice to these products.

So what are the lessons learned?

Nothing is inherently wrong with securitization itself. It can improve the flow and circulation of credit, and it can allow more investors to participate in the market by segmenting their desired risk and return. However, what securitization does is to help one gain an exposure to the underlying assets, so you get out of it what you put into it. We can have a debate about what is the social purpose of finer and finer tranching exercises, but if you think that one leg has value, then we also need to recognize we are arguing over a line that no one can observe and that is highly subjective.

Which means that securitization is indeed an important part of the credit creation process. Therefore we need to improve oversight at the front end of the chain when we're deciding what to put into a securitization.

We never should have been complacent with the laissez-faire policy of allowing mortgage brokers and aggregators to run amok. Although we thought the consumer was best served by competition, we took our hands off the reins of credit creation. As a result, we couldn't alter margin requirements or lending standards even if we had wanted to. (And as Greenspan pointed out in Congressional testimony, that wouldn't have been a politically popular decision to begin with, as politicians were pushing for more and more access for the little guy.)

We also need to think about evaluating our tax policies of promoting debt over equity. Do we need to subsidize borrowing? As mentioned in a previous piece, leverage leads to volatility, volatility to instability. Do we need to encourage this as part of national policy?

Lastly, models can be useful, but they should not replace judgment. We must realize that our assumptions about how the world works can be wrong. Just as the rating agencies models had prebuilt biases, so did many of the VAR models of financial institutions. And those VAR models had assumptions on products that had assumptions. People too reliant on models will both live and die by them. We just need to remember the "die" part.

Eric Oberg worked in fixed income, currencies and commodities for Goldman Sachs for 17 years before retiring as a managing director.