Over the last two years of the housing bubble, Wall Street bankers perpetrated one of the greatest episodes of self-dealing in financial history.

Faced with increasing difficulty in selling the mortgage-backed securities that had been among their most lucrative products, the banks hit on a solution that preserved their quarterly earnings and huge bonuses:

They created fake demand.

A ProPublica analysis shows for the first time the extent to which banks -- primarily Merrill Lynch, but also Citigroup, UBS and others -- bought their own products and cranked up an assembly line that otherwise should have flagged.

The products they were buying and selling were at the heart of the 2008 meltdown -- collections of mortgage bonds known as collateralized debt obligations, or CDOs.

As the housing boom began to slow in mid-2006, investors became skittish about the riskier parts of those investments. So the banks created -- and ultimately provided most of the money for -- new CDOs. Those new CDOs bought the hard-to-sell pieces of the original CDOs. The result was a daisy chain that solved one problem but created another: Each new CDO had its own risky pieces. Banks created yet other CDOs to buy those.

Individual instances of these questionable trades have been reported before, but ProPublica's investigation, done in partnership with NPR's Planet Money, shows that by late 2006 they became a common industry practice.

An analysis by research firm Thetica Systems, commissioned by ProPublica, shows that in the last years of the boom, CDOs had become the dominant purchaser of key, risky parts of other CDOs, largely replacing real investors like pension funds. By 2007, 67 percent of those slices were bought by other CDOs, up from 36 percent just three years earlier. The banks often orchestrated these purchases. In the last two years of the boom, nearly half of all CDOs sponsored by market leader Merrill Lynch bought significant portions of other Merrill CDOs.

ProPublica also found 85 instances during 2006 and 2007 in which two CDOs bought pieces of each other. These trades, which involved $107 billion worth of CDOs, underscore the extent to which the market lacked real buyers. Often the CDOs that swapped purchases closed within days of each other, the analysis shows.

There were supposed to be protections against this sort of abuse. While banks provided the blueprint for the CDOs and marketed them, they typically selected independent managers who chose the specific bonds to go inside them. The managers had a legal obligation to do what was best for the CDO. They were paid by the CDO, not the bank, and were supposed to serve as a bulwark against self-dealing by the banks, which had the fullest understanding of the complex and lightly regulated mortgage bonds.

It rarely worked out that way. The managers were beholden to the banks that sent them the business. On a billion-dollar deal, managers could earn a million dollars in fees, with little risk. Some small firms did several billion dollars of CDOs in a matter of months.

"All these banks for years were spawning trading partners," says a former executive from Financial Guaranty Insurance Company, a major insurer of the CDO market. "You don't have a trading partner? Create one."

The executive, like most of the dozens of people ProPublica spoke with about the inner workings of the market at the time, asked not to be named out of fear of being sucked into ongoing investigations or because they are involved in civil litigation.

Keeping the assembly line going had a wealth of short-term advantages for the banks. Fees rolled in. A typical CDO could net the bank that created it between $5 million and $10 million -- about half of which usually ended up as employee bonuses. Indeed, Wall Street awarded record bonuses in 2006, a hefty chunk of which came from the CDO business.

The self-dealing super-charged the market for CDOs, enticing some less-savvy investors to try their luck. Crucially, such deals maintained the value of mortgage bonds at a time when the lack of buyers should have driven their prices down.

But the strategy of speeding up the assembly line had devastating consequences for homeowners, the banks themselves and, ultimately, the global economy. Because of Wall Street's machinations, more mortgages had been granted to ever-shakier borrowers. The results can now be seen in foreclosed houses across America.

The incestuous trading also made the CDOs more intertwined and thus fragile, accelerating their decline in value that began in the fall of 2007 and deepened over the next year. Most are now worth pennies on the dollar. Nearly half of the nearly trillion dollars in losses to the global banking system came from CDOs, losses ultimately absorbed by taxpayers and investors around the world. The banks' troubles sent the world's economies into a tailspin from which they have yet to recover.

It remains unclear whether any of this violated laws. The SEC has said that it is actively looking at as many as 50 CDO managers as part of its broad examination of the CDO business' role in the financial crisis. In particular, the agency is focusing on the relationship between the banks and the managers. The SEC is exploring how deals were structured, if any quid pro quo arrangements existed, and whether banks pressured managers to take bad assets.

The banks declined to directly address ProPublica's questions. Asked about its relationship with managers and the cross-ownership among its CDOs, Citibank responded with a one-sentence statement:

"It has been widely reported that there are ongoing industry-wide investigations into CDO-related matters and we do not comment on pending investigations."

None of ProPublica's questions had mentioned the SEC or pending investigations.

Posed a similar list of questions, Bank of America, which now owns Merrill Lynch, said:

"These are very specific questions regarding individuals who left Merrill Lynch several years ago and a CDO origination business that, due to market conditions, was discontinued by Merrill before Bank of America acquired the company."

This is the second installment of a ProPublica series about the largely hidden history of the CDO boom and bust. Our first story looked at how one hedge fund helped create at least $40 billion in CDOs as part of a strategy to bet against the market. This story turns the focus on the banks.

Merrill Lynch Pioneers Pervert the Market

By 2004, the housing market was in full swing, and Wall Street bankers flocked to the CDO frenzy. It seemed to be the perfect money machine, and for a time everyone was happy.

Homeowners got easy mortgages. Banks and mortgage companies felt secure lending the money because they could sell the mortgages almost immediately to Wall Street and get back all their cash plus a little extra for their trouble. The investment banks charged massive fees for repackaging the mortgages into fancy financial products. Investors all around the world got to play in the then-phenomenal American housing market.

The mortgages were bundled into bonds, which were in turn combined into CDOs offering varying interest rates and levels of risk.

Investors holding the top tier of a CDO were first in line to get money coming from mortgages. By 2006, some banks often kept this layer, which credit agencies blessed with their highest rating of Triple A.

Buyers of the lower tiers took on more risk and got higher returns. They would be the first to take the hit if homeowners funding the CDO stopped paying their mortgages. (Here's a video explaining how CDOs worked.)

Over time, these risky slices became increasingly hard to sell, posing a problem for the banks. If they remained unsold, the sketchy assets stayed on their books, like rotting inventory. That would require the banks to set aside money to cover any losses. Banks hate doing that because it means the money can't be loaned out or put to other uses.

Being stuck with the risky portions of CDOs would ultimately lower profits and endanger the whole assembly line.

The banks, notably Merrill and Citibank, solved this problem by greatly expanding what had been a common and accepted practice: CDOs buying small pieces of other CDOs.

Architects of CDOs typically included what they called a "bucket" -- which held bits of other CDOs paying higher rates of interest. The idea was to boost overall returns of deals primarily composed of safer assets. In the early days, the bucket was a small portion of an overall CDO.

One pioneer of pushing CDOs to buy CDOs was Merrill Lynch's Chris Ricciardi, who had been brought to the firm in 2003 to take Merrill to the top of the CDO business. According to former colleagues, Ricciardi's team cultivated managers, especially smaller firms.

Merrill exercised its leverage over the managers. A strong relationship with Merrill could be the difference between a business that thrived and one that didn't. The more deals the banks gave a manager, the more money the manager got paid.

As the head of Merrill's CDO business, Ricciardi also wooed managers with golf outings and dinners. One Merrill executive summed up the overall arrangement: "I'm going to make you rich. You just have to be my bitch."

But not all managers went for it.

An executive from Trainer Wortham, a CDO manager, recalls a 2005 conversation with Ricciardi. "I wasn't going to buy other CDOs. Chris said: 'You don't get it. You have got to buy other guys' CDOs to get your deal done. That's how it works.'" When the manager refused, Ricciardi told him, "'That's it. You are not going to get another deal done.'" Trainer Wortham largely withdrew from the market, concerned about the practice and the overheated prices for CDOs.

Ricciardi declined multiple requests to comment.

Merrill CDOs often bought slices of other Merrill deals. This seems to have happened more in the second half of any given year, according to ProPublica's analysis, though the purchases were still a small portion compared to what would come later. Annual bonuses are based on the deals bankers completed by yearend.

Ricciardi left Merrill Lynch in February 2006. But the machine he put into place not only survived his departure, it became a model for competitors.

As Housing Market Wanes, Self-Dealing Takes Off

By mid-2006, the housing market was on the wane. This was particularly true for subprime mortgages, which were given to borrowers with spotty credit at higher interest rates. Subprime lenders began to fold, in what would become a mass extinction. In the first half of the year, the percentage of subprime borrowers who didn't even make the first month's mortgage payment tripled from the previous year.

That made CDO investors like pension funds and insurance companies increasingly nervous. If homeowners couldn't make their mortgage payments, then the stream of cash to CDOs would dry up. Real "buyers began to shrivel and shrivel," says Fiachra O'Driscoll, who co-ran Credit Suisse's CDO business from 2003 to 2008.

Faced with disappearing investor demand, bankers could have wound down the lucrative business and moved on. That's the way a market is supposed to work. Demand disappears; supply follows. But bankers were making lots of money. And they had amassed warehouses full of CDOs and other mortgage-based assets whose value was going down.

Rather than stop, bankers at Merrill, Citi, UBS and elsewhere kept making CDOs.

The question was: Who would buy them?

The top 80 percent, the less risky layers or so-called "super senior," were held by the banks themselves. The beauty of owning that supposedly safe top portion was that it required hardly any money be held in reserve.

That left 20 percent, which the banks did not want to keep because it was riskier and required them to set aside reserves to cover any losses. Banks often sold the bottom, riskiest part to hedge funds. That left the middle layer, known on Wall Street as the "mezzanine," which was sold to new CDOs whose top 80 percent was ultimately owned by ... the banks.

"As we got further into 2006, the mezzanine was going into other CDOs," says Credit Suisse's O'Driscoll.

This was the daisy chain. On paper, the risky stuff was gone, held by new independent CDOs. In reality, however, the banks were buying their own otherwise unsellable assets.

How could something so seemingly short-sighted have happened?

It's one of the great mysteries of the crash. Banks have fleets of risk managers to defend against just such reckless behavior. Top executives have maintained that while they suspected that the housing market was cooling, they never imagined the crash. For those doing the deals, the payoff was immediate. The dangers seemed abstract and remote.

The CDO managers played a crucial role. CDOs were so complex that even buyers had a hard time seeing exactly what was in them -- making a neutral third party that much more essential.

"When you're investing in a CDO you are very much putting your faith in the manager," says Peter Nowell, a former London-based investor for the Royal Bank of Scotland. "The manager is choosing all the bonds that go into the CDO." (RBS suffered mightily in the global financial meltdown, posting the largest loss in United Kingdom history, and was de facto nationalized by the British government.)

By persuading managers to pick the unsold slices of CDOs, the banks helped keep the market going. "It guaranteed distribution when, quite frankly, there was not a huge market for them," says Nowell.

The counterintuitive result was that even as investors began to vanish, the mortgage CDO market more than doubled from 2005 to 2006, reaching $226 billion, according to the trade publication Asset-Backed Alert.

Citi and Merrill Hand Out Sweetheart Deals

As the CDO market grew, so did the number of CDO management firms, including many small shops that relied on a single bank for most of their business. According to Fitch, the number of CDO managers it rated rose from 89 in July 2006 to 140 in September 2007.

One CDO manager epitomized the devolution of the business, according to numerous industry insiders: a Wall Street veteran named Wing Chau.

Earlier in the decade, Chau had run the CDO department for Maxim Group, a boutique investment firm in New York. Chau had built a profitable business for Maxim based largely on his relationship with Merrill Lynch. In just a few years, Maxim had corralled more than $4 billion worth of assets under management just from Merrill CDOs.

In August 2006, Chau bolted from Maxim to start his own CDO management business, taking several colleagues with him. Chau's departure gave Merrill, the biggest CDO producer, one more avenue for unsold inventory.

Chau named the firm Harding, after the town in New Jersey where he lived. The CDO market was starting its most profitable stretch ever, and Harding would play a big part. In an eleven-month period, ending in August 2007, Harding managed $13 billion of CDOs, including more than $5 billion from Merrill, and another nearly $5 billion from Citigroup. (Chau would later earn a measure of notoriety for a cameo appearance in Michael Lewis' bestseller "The Big Short," where he is depicted as a cheerfully feckless "go-to buyer" for Merrill Lynch's CDO machine.)

Chau had a long-standing friendship with Ken Margolis, who was Merrill's top CDO salesman under Ricciardi. When Ricciardi left Merrill in 2006, Margolis became a co-head of Merrill's CDO group. He carried a genial, let's-just-get-the-deal-done demeanor into his new position. An avid poker player, Margolis told a friend that in a previous job he had stood down a casino owner during a foreclosure negotiation after the owner had threatened to put a fork through his eye.

Chau's close relationship with Merrill continued. In late 2006, Merrill sublet office space to Chau's startup in the Merrill tower in Lower Manhattan's financial district. A Merrill banker, David Moffitt, scheduled visits to Harding for prospective investors in the bank's CDOs. "It was a nice office," overlooking New York Harbor, recalls a CDO buyer. "But it did feel a little weird that it was Merrill's building," he said.

Moffitt did not respond to requests for comment.

Under Margolis, other small managers with meager track records were also suddenly handling CDOs valued at as much as $2 billion. Margolis declined to answer any questions about his own involvement in these matters.

A Wall Street Journal article ($) from late 2007, one of the first of its kind, described how Margolis worked with one inexperienced CDO manager called NIR on a CDO named Norma, in the spring of that year. The Long Island-based NIR made about $1.5 million a year for managing Norma, a CDO that imploded.

"NIR's collateral management business had arisen from efforts by Merrill Lynch to assemble a stable of captive small firms to manage its CDOs that would be beholden to Merrill Lynch on account of the business it funneled to them," alleged a lawsuit filed in New York state court against Merrill over Norma that was settled quietly after the plaintiffs received internal Merrill documents.

NIR declined to comment.

Banks had a variety of ways to influence managers' behavior.

Some of the few outside investors remaining in the market believed that the manager would do a better job if he owned a small slice of the CDO he was managing. That way, the manager would have more incentive to manage the investment well, since he, too, was an investor. But small management firms rarely had money to invest. Some banks solved this problem by advancing money to managers such as Harding.

Chau's group managed two Citigroup CDOs -- 888 Tactical Fund and Jupiter High-Grade VII -- in which the bank loaned Harding money to buy risky pieces of the deal. The loans would be paid back out of the fees the managers took from the CDO and its investors. The loans were disclosed to investors in a few sentences among the hundreds of pages of legalese accompanying the deals.

In response to ProPublica's questions, Chau's lawyer said, "Harding Advisory's dealings with investment banks were proper and fully disclosed."

Citigroup made similar deals with other managers. The bank lent money to a manager called Vanderbilt Capital Advisors for its Armitage CDO, completed in March 2007.

Vanderbilt declined to comment. It couldn't be learned how much money Citigroup loaned or whether it was ever repaid.

Yet again banks had masked their true stakes in CDO. Banks were lending money to CDO managers so they could buy the banks' dodgy assets. If the managers couldn't pay the loans back -- and most were thinly capitalized -- the banks were on the hook for even more losses when the CDO business collapsed.

Goldman, Merrill and Others Get Tough

When the housing market deteriorated, banks took advantage of a little-used power they had over managers.

The way CDOs are put together, there is a brief period when the bonds picked by managers sit on the banks' balance sheets. Because the value of such assets can fall, banks reserved the right to overrule managers' selections.

According to numerous bankers, managers and investors, banks rarely wielded that veto until late 2006, after which it became common. Merrill was in the lead.

"I would go to Merrill and tell them that I wanted to buy, say, a Citi bond," recalls a CDO manager. "They would say 'no.' I would suggest a UBS bond, they would say 'no.' Eventually, you got the joke." Managers could choose assets to put into their CDOs but they had to come from Merrill CDOs. One rival investment banker says Merrill treated CDO managers the way Henry Ford treated his Model T customers: You can have any color you want, as long as it's black.

Once, Merrill's Ken Margolis pushed a manager to buy a CDO slice for a Merrill-produced CDO called Port Jackson that was completed in the beginning of 2007: "'You don't have to buy the deal but you are crazy if you don't because of your business,'" an executive at the management firm recalls Margolis telling him. "'We have a big pipeline and only so many more mandates to give you.' You got the message." In other words: Take our stuff and we'll send you more business. If not, forget it.

Margolis declined to comment on the incident.

"All the managers complained about it," recalls O'Driscoll, the former Credit Suisse banker who competed with other investment banks to put deals together and market them. But "they were indentured slaves." O'Driscoll recalls managers grumbling that Merrill in particular told them "what to buy and when to buy it."

Other big CDO-producing banks quickly adopted the practice.

A little-noticed document released this year during a congressional investigation into Goldman Sachs' CDO business reveals that bank's thinking. The firm wrote a November 2006 internal memorandum about a CDO called Timberwolf, managed by Greywolf, a small manager headed by ex-Goldman bankers. In a section headed "Reasons To Pursue," the authors touted that "Goldman is approving every asset" that will end up in the CDO. What the bank intended to do with that approval power is clear from the memo: "We expect that a significant portion of the portfolio by closing will come from Goldman's offerings."

When asked to comment whether Goldman's memo demonstrates that it had effective control over the asset selection process and that Greywolf was not in fact an independent manager, the bank responded: "Greywolf was an experienced, independent manager and made its own decisions about what reference assets to include. The securities included in Timberwolf were fully disclosed to the professional investors who invested in the transaction."

Greywolf declined to comment. One of the investors, Basis Capital of Australia, filed a civil lawsuit in federal court in Manhattan against Goldman over the deal. The bank maintains the lawsuit is without merit.

By March 2007, the housing market's signals were flashing red. Existing home sales plunged at the fastest rate in almost 20 years. Foreclosures were on the rise. And yet, to CDO buyer Peter Nowell's surprise, banks continued to churn out CDOs.

"We were pulling back. We couldn't find anything safe enough," says Nowell. "We were amazed that April through June they were still printing deals. We thought things were over."

Instead, the CDO machine was in overdrive. Wall Street produced $70 billion in mortgage CDOs in the first quarter of the year.

Many shareholder lawsuits battling their way through the court system today focus on this period of the CDO market. They allege that the banks were using the sales of CDOs to other CDOs to prop up prices and hide their losses.

"Citi's CDO operations during late 2006 and 2007 functioned largely to sell CDOs to yet newer CDOs created by Citi to house them," charges a pending shareholder lawsuit against the bank that was filed in federal court in Manhattan in February 2009. "Citigroup concocted a scheme whereby it repackaged many of these investments into other freshly-baked vehicles to avoid incurring a loss."

Citigroup described the allegations as "irrational," saying the bank's executives would never knowingly take actions that would lead to "catastrophic losses."

In the Hall of Mirrors, Myopic Rating Agencies

The portion of CDOs owned by other CDOs grew right alongside the market. What had been 5 percent of CDOs (remember the "bucket") now came to constitute as much as 30 or 40 percent of new CDOs. (Wall Street also rolled out CDOs that were almost entirely made up of CDOs, called CDO squareds.)

The ever-expanding bucket provided new opportunities for incestuous trades.

It worked like this: A CDO would buy a piece of another CDO, which then returned the favor. The transactions moved both CDOs closer to completion, when bankers and managers would receive their fees.

ProPublica's analysis shows that in the final two years of the business, CDOs with cross-ownership amounted to about one-fifth of the market, about $107 billion.

Here's an example from early May 2007:

• A CDO called Jupiter VI bought a piece of a CDO called Tazlina II.
• Tazlina II bought a piece of Jupiter VI.

Both Jupiter VI and Tazlina II were created by Merrill and were completed within a week of each other. Both were managed by small firms that did significant business with Merrill: Jupiter by Wing Chau's Harding, and Tazlina by Terwin Advisors. Chau did not respond to questions about this deal. Terwin Advisors could not reached.

Just a few weeks earlier, CDO managers completed a comparable swap between Jupiter VI and another Merrill CDO called Forge 1.

Forge has its own intriguing history. It was the only deal done by a tiny manager of the same name based in Tampa, Fla. The firm was started less than a year earlier by several former Wall Street executives with mortgage experience. It received seed money from Bryan Zwan, who in 2001 settled an SEC civil lawsuit over his company's accounting problems in a federal court in Florida. Zwan and Forge executives didn't respond to requests for comment.

After seemingly coming out of nowhere, Forge won the right to manage a $1.5 billion Merrill CDO. That earned Forge a visit from the rating agency Moody's.

"We just wanted to make sure that they actually existed," says a former Moody's executive. The rating agency saw that the group had an office near the airport and expertise to do the job.

Rating agencies regularly did such research on managers, but failed to ask more fundamental questions. The credit ratings agencies "did heavy, heavy due diligence on managers but they were looking for the wrong things: how you processed a ticket or how your surveillance systems worked," says an executive at a CDO manager. "They didn't check whether you were buying good bonds."

One Forge employee recalled in a recent interview that he was amazed Merrill had been able to find buyers so quickly. "They were able to sell all the tranches" -- slices of the CDO -- "in a fairly rapid period of time," said Rod Jensen, a former research analyst for Forge.

Forge achieved this feat because Merrill sold the slices to other CDOs, many linked to Merrill.

The ProPublica analysis shows that two Merrill CDOs, Maxim II and West Trade III, each bought pieces of Forge. Small managers oversaw both deals.

Forge, in turn, was filled with detritus from Merrill. Eighty-two percent of the CDO bonds owned by Forge came from other Merrill deals.

Citigroup did its own version of the shuffle, as these three CDOs demonstrate:

• A CDO called Octonion bought some of Adams Square Funding II.
• Adams Square II bought a piece of Octonion.
• A third CDO, Class V Funding III, also bought some of Octonion.
• Octonion, in turn, bought a piece of Class V Funding III.

All of these Citi deals were completed within days of each other. Wing Chau was once again a central player. His firm managed Octonion. The other two were managed by a unit of Credit Suisse. Credit Suisse declined to comment.

Not all cross-ownership deals were consummated.

In spring 2007, Deutsche Bank was creating a CDO and found a manager that wanted to take a piece of it. The manager was overseeing a CDO that Merrill was assembling. Merrill blocked the manager from putting the Deutsche bonds into the Merrill CDO. A former Deutsche Bank banker says that when Deutsche Bank complained to Andy Phelps, a Merrill CDO executive, Phelps offered a quid pro quo: If Deutsche was willing to have the manager of its CDO buy some Merrill bonds, Merrill would stop blocking the purchase. Phelps declined to comment.

The Deutsche banker, who says its managers were independent, recalls being shocked: "We said we don't control what people buy in their deals." The swap didn't happen.

The Missing Regulators and the Aftermath

In September 2007, as the market finally started to catch up with Merrill Lynch, Ken Margolis left the firm to join Wing Chau at Harding.

Chau and Margolis circulated a marketing plan for a new hedge fund to prospective investors touting their expertise in how CDOs were made and what was in them. The fund proposed to buy failed CDOs -- at bargain basement prices. In the end, Margolis and Chau couldn't make the business work and dropped the idea.

Why didn't regulators intervene during the boom to stop the self-dealing that had permeated the CDO market?

No one agency had authority over the whole business. Since the business came and went in just a few years, it may have been too much to expect even assertive regulators to comprehend what was happening in time to stop it.

While the financial regulatory bill passed by Congress in July creates more oversight powers, it's unclear whether regulators have sufficient tools to prevent a replay of the debacle.

In just two years, the CDO market had cut a swath of destruction. Partly because CDOs had bought so many pieces of each other, they collapsed in unison. Merrill Lynch and Citigroup, the biggest perpetrators of the self-dealing, were among the biggest losers. Merrill lost about $26 billion on mortgage CDOs and Citigroup about $34 billion.

Additional reporting by Kitty Bennett, Krista Kjellman Schmidt, Lisa Schwartz and Karen Weise.

An interactive graphic we published includes at least one example of cross-exposure that did not involve "unsold" inventory. A CDO called Tourmaline III made a sidebet in 2007 that mirrored the performance of a piece of a CDO called Zais Investment Grade 8; that same year Zais 8 bought a piece of Tourmaline III. Both CDOs were underwritten by Deutsche Bank.