Arnold Kling offers a very concise view of the financial intermediation:

[T]he nonfinancial sector would love to issue risky long-term liabilities to fund investments, while consumers would love to hold riskless short-term assets to maintain liquidity. The financial sector intermediates by holding risky long-term assets and issuing riskless short-term liabilities. The more the financial sector expands, the more long-term investment is undertaken in the economy.

In order for the financial sector to do its job properly, it needs to enjoy the right amount of confidence from the public. With too little confidence, the economy suffers from credit scarcity and insufficient long-term investment. With too much confidence, the economy suffers from bubbles and excess credit creation, followed by crashes.

I think this is an eloquent statement of a very common view. It also beautifully isolates the problem with how we have constructed financial intermediation.

I am certain it is true that the nonfinacial sector would love to hold short-term risk-free assets, especially if they pay a high return. It is also true that businesses that invest in real projects and seek to minimize financial risk would prefer to issue long-term liabilities that try to match payouts to lenders with expected project cashflows. However, this gap is not, in fact, intermediable. An intermediary that claims to offer truly riskless assets against investments in risky projects must rely upon subsidy, subterfuge, or both.

An intermediary can "add value" by reducing investors' risk in comparison to disintermediated investment, by for example, investing in a better-diversified portfolio than an investor would. An intermediary can very effectively reduce liquidity risks to investors, again by since idiosyncratic liquidity demands are themselves diversifiable. But risk reduction via these techniques can never reduce risk to zero. In fact, investing via an intermediary can never alter the fact that 100% of invested capital is at risk — business performance is not uncorrelated and projects can fail completely. Also, usually idiosyncratic liquidity demand occasionally become highly correlated, due to bank runs or real need for cash. Statistical attempts to quantify these risks are misleading at best, as the distributions from which inferences are drawn are violently nonstationary — the world is always changing, the past is never a great guide to the future for very long. Fundamentally, the value intermediaries can add by diversifying over investments and liquidity requirements is very modest, and ought to be acknowledged as such.

Furthermore, if diversifying across operation and liquidity risk was the value provided by financial intermediaries, they should have largely been competed out of the business, as investors can buy and sell diverse portfolios of liquid securities as easily as investing in bank deposits, either constructing them directly or purchasing diverse vehicles, like ETFs or ABS.

So, what is the "value-add" of the financial sector? Let's go back to Arnold's think-nugget. Investors cannot, on their own, create short-term truly riskless assets that pass through the returns of risky and illiquid busines projects. So, we can see why there's demand for banks: They do give the people what they want, on both sides of the funding equation. But, as we've already seen, in reality they can't give the people what they want without subsidy or subterfuge. Either some truly riskless guarantee has to backstop both liquidity and solvency risk, or intermediaries have to lie and pretend that their assets are riskless.

Note that insurance is an insufficient means of squaring this circle. If a bank purchases private sector deposit insurance or liquidity commitments, that is an asset purchase that may reduce its overall portfolio risk, but that cannot eliminate it, especially during times when correlations run high. A private sector insurance policy is a risky asset. Governments can offer riskless liquidity insurance and insurance against the nominal (but not real) value of financial sector liabilities. But it cannot do so at "a market rate". Government is structurally the monopoly seller of this form of insurance for assets denominated in the currency that it issues. Further, the risks it must insure cannot be actuarially priced without heroic assumptions — the distribution of systemic risk is nonstationary, the future is everchanging and extrapolations always break eventually. (It's fair to say, however, that zero is too low a price.)

So how does the financial sector seem to offer risk-free assets against risky projects? I think "constructive ambiguity" is the right phrase. The government provides subsidies in the form of literally priceless deposit and liquidity backstops, but those are explicitly limited. Banks work to diligently to increase both the level of insurance and degree to which assets are perceived to be insured by becoming so large that social costs of a bank default on even notionally risky assets are thought to exceed the costs to government of paying out on insurance policies to which it never agreed. Even a very careful observer cannot tell a priori whether many assets offered are genuinely riskless or not, that is to what degree the risk-free status of bank assets is due to subsidy, and to what degree to subterfuge. But there is an ingenious tinkerbell aspect to the risk status of bank assets: If, with a bit of subterfuge, risky assets can be sold as riskless assets, then the social costs of default rise, since asset holders will not have privately managed the risk that the asset might fail. The increase in social costs created by a mischaracterization of a risky asset as riskless, however, alters the likelihood that an asset will be de facto insured. There is a game theoretic equilibrium, that works to the advantage of intermediaries and their customers on both sides of the funding stream, whereby banks offer assets in large quantities as though they are risk-free, and investors accept and treat those assets as risk-free, and by believing together in what is formally not true, they create costs to the sovereign so large if it is not true that the sovereign makes it true. This is an equilibrium, a predictable outcome, not an aberration. And it does happen all the time.

In theory, this is a repeated game, and governments might eliminate the bad equilibrium by committing to bearing social costs that are higher than the immediate costs of providing ex post insurance, in order to prevent the subsidy-extracting equilibrium from taking hold. That's what people who'd like to see a lot of banks go bust due to "moral hazard" concerns think should happen. But, if governments are unable to credibly commit to accepting large social costs, investors, borrowers, and intermediaries will test them by trying to extract the subsidy of infinite insurance. In practice, it's clear that governments have rarely been able to credibly commit to stick to the letter of their limited insurance commitments.

Alternatively, governments can accept the extraction of a de facto insurance subsidy, but supervise intermediaries to try to mitigate the cost of future claims. But that's a difficult task, as all private sector actors, borrowers, lenders, and intermediaries, have an incentive to maximize the subsidy extracted from the state, and will collude in creative ways to do so. (Of course, eventually these actors pay taxes, but even if they are sufficiently far-sighted to consider that, the distribution of benefits from the extracted subsidy is not coincident with the distribution of expected tax liabilities or inflation costs.) As long as it is possible to create a situation where the social costs of failure imply a bail-out, creative financiers will work to capture the huge value of what is essentially an option on an entire macroeconomy (even a global economy).

So, what is to be done? Here are some suggestions:

1. Eliminate the "constructive ambiguity" that permits private sector actors to offer apparently risk-free, instantaneously redeemable securities. Eliminate government insurance of deposits and all other assets except for direct obligations of the state. This is insufficient — AIG, for example, has extracted a very large bailout despite having never offered insured deposits, and there have been bank bailouts since long before formal deposit insurance. But banks' ability to muddy the waters between explicitly guaranteed and other assets seems to facilitate the process by which investors naively or cynically confuse risky for risk-free assets. Banking crises are larger and more frequent than any other sort of financial crisis that compels a state subsidy.

2. Keep financial firms small enough and sufficiently well compartmentalized from one another that a failure of one or several does not create social costs large enough to force a sovereign payout. Discourage portfolio correlation by creating a fiduciary obligation of independent evaluation that places agents who copycat or herd in jeopardy of investor lawsuits.

3. When possible, disintermediate. Nonfinancial firms are subject to clear operational risks, which forces direct investors to manage risks privately. Risk-management by private investors reduces the social cost of failure that compels government bailouts. Nonfinacial firms sometimes succeed at extracting bailouts, but much more rarely and with a larger fraction of the costs borne by investors than when financial firms do so.
4. Prefer equity to debt arrangements. All business risks must eventually be borne by investors. Debt financing concentrates enterprise risk among equity holders, and enables debt-holders to manage their investment risk less carefully. Risk-management by private investors reduces the social cost of failure that compels government bailouts.
5. Vaccinate investors by maintaining some level of asset price risk. Governments should limit liquidity and promote short-term price volatility of risky assets, perhaps quite artificially, to ensure that investors are always provisioning to manage risk. Risk-management by private investors reduces the social cost of failure that compels government bailouts.
6. Carefully ration risk-free obligations offered by the state. As Winterspeak likes to point out, states do not need to issue debt to fund themselves in their own currency. States can more equitably and transparently promote price stability via taxation than by borrowing and intervening in sprawling debt markets. Government debt ought not be viewed as loans to fund operations, but as a form of insurance offered by the state to private savers in strictly limited quantity, in order that people of modest means don't have to perform the work and bear the risk of managing uncertain investments. Governments should strive to guarantee a near-zero, but always positive, real return on these obligations, permitting risk-averse savers to transport purchasing power into the future, but not magnify real wealth. Those who want a high return must do the work and bear the risk of achieving it. The state should not guarantee that.

I know that the political economy of these suggestions is, well, iffy. But one should never underestimate how much political economy considerations might change during the turbulence of a global financial crisis.