Correction: With regard to weakness (2), I should have clearly written that, in the case of an ETH price increase, the algorithm incentivizes the creation of more DAI, and thus, the locking of additional ETH.
Before we proceed, two caveats are immediately in order:
- Sometimes, things that do not work in theory turn out to work in practice because the world is complex and theorists may not take into account all the relevant possibilities. Cryptocurrencies have excelled at proving theorists wrong  and I will be more than happy to end up being yet another annoying ivory tower dweller here.
- I have only read the DAI white paper and the simplified explanation but not the purple paper, thus I may have misunderstood or not taken into account certain important technical details.
The DAI is an ingenuous project aimed at creating a stablecoin, i.e. a cryptographic asset whose volatility will be low compared to other cryptocurrencies like BTC and ETH. It attempts to achieve this through targeting a 1 Dai = 1$ exchange rate using a smart contract-based collateral mechanism involving ETH. For more detail on how the scheme is supposed to work see the sources I mentioned above.
There are two important problems I see with the project from the economic perspective, the latter one being potentially a lot more serious: the higher than one-to-one collateralization and the adjustments to collateral price movements creating runaway feedback loops.
- The higher than one-to-one collateralization
Each unit of DAI is supposed to be backed by an amount of ETH that at the moment of the unit’s creation is worth more than 1$. The reason for this is that were it not the case, DAI would just be backed by a wildly fluctuating asset and would thus provide no stability. Instead, the DAI scheme provides for adjustment in the amount of DAI issued before the value of collateral per unit of DAI plunges below $1.
This solution, however, creates the risk that people will buy DAI only to effectively acquire the underlying ETH at a non-DAI-denominated price which is lower than at the market for ETH. The DAI scheme involves a complicated mechanism involving another token (Maker) probably aimed at preventing this (even though this exact risk is not explicitly acknowledged) but it is far from clear if it is sufficient.
Even if the stability fee mechanism is adequate, it creates the problem that, in contrast to the historical gold standard, there is no on-demand convertibility of DAI into ETH, which may hurt adoption. How important a problem this may become is hard to predict, though.
2. Price adjustments further affecting prices
A potentially much more serious problem with DAI arises from the way the supply of DAI is supposed to adjust to the collateral price volatility. If the ETH price falls significantly, a complex algorithm leads to burning of some DAI to restore the peg. In the reverse scenario, users are incentivized to create more DAI.
Let us make a big assumption that the mechanisms in question work as advertized. There is still the problem that if the DAI in circulation are secured by a substantial proportion of the existing ETH, the releases of ETH into the market in the case of adjustments to ETH price decreases or the withdrawals of additional ETH from circulation to accommodate price increases may themselves substantially affect the ETH price and create self-reinforcing loops.
Consider a hypothetical example. Suppose the DAI is mature enough that its supply is at some point backed by 20% of ETH in existence. Suppose the price of ETH rises by 10%. Suppose that, to accommodate this, the DAI algorithm incentivizes users to increase the ETH collateral by 10%, which translates into additional 2% of ETH in existence being locked. This reduces the ETH supply which may translate into a further ETH price increase and the need for further need for upward DAI supply and collateral adjustment.
One might respond that further adjustments would require locking less and less ETH, until ETH price doesn’t respond to asymptotic adjustments anymore. However, the flaw in this reasoning is that small increases or decreases in supply may generate disproportionately large price movements in certain conditions. Just consider the massive fluctuations in the oil price. They are caused by much smaller differences between supply and demand. Add to this that people may start to anticipate how the ETH price may respond to the DAI adjustments, and the adjustment algorithm may result in chaos.
A possible solution to this problem would involve using several types of collateral, and ajusting by locking or releasing small amounts of each. However, the operations with several cryptographic assets are not costless, and this may well render the stablecoin system too expensive to use.
Are stablecoins really necessary for widespread cryptocurrency adoption?
This naturally leads to the question whether stablecoins are truly necessary for mass cryptocurrency adoption. While they would probably facilitate it, this does not seem to be the case, especially if we are talking about coins like ETH rather than the purely transactional ones like BTC, DASH or Monero.
The problem with the mass adoption of the latter is ultimately the one Mises imagined when he formulated the regression theorem of the monetary value. In order to start using some commodity as a means of exchange for all kinds of goods, someone needs to believe that it will retain certain value over time. This value may not just be created out of thin air. It may be argued that coins like BTC may have such intrinsic value to their owners, and this is partly true, given that BTC has in a limited sense become a means of exchange. However, for mass adoption, this may not be enough.
Enter cryptocurrencies like ETH or even tokens like Augur’s REP.  Their key feature is that certain blockchain-based services that might potentially become widely used are tied to them. You cannot use the smart contracts on the Ethereum blockchain without paying in ETH, for example. This may provide the very initial non-monetary value basis a genuine money may necessarily have to have.
 First, computer scientists had believed that distributed consensus problems like the Byzantine generals problem could not be solved but then came Satoshi Nakamoto and provided a solution for a special case involving a virtual currency used to reward the verifiers. Later, a few economists like Kevin Dowd claimed that PoW currencies like Bitcoin are inherently contradictory in that they rely on mining that resembles a natural monopoly (it is cheaper to mine if all mining is concentrated in a single entity since difficulty increases with increasing hashpower).
 For those, who see market manipulation on every turn, I don’t hold REP or any significant amount of ETH, and use them as mere examples.