Unsystematic Risk: It’s said that diversification eliminates (or just mitigates?) unsystematic risk. But is this absolute? If an investor still has exposure to a company or industry (e.g., through concentrated holdings), is the risk merely reduced rather than eliminated? Wouldn’t full elimination require zero exposure?
Perfect diversification eliminates unsystematic risk (assuming Efficient Market Hypothesis; and if you aren't assuming EMH, then all bets are off as to what is optimal allocation). That follows from the definitions of "perfect diversification" and "unsystematic risk". Perfect diversification is diversification the minimizes risk for a particular level of return, systematic risk is what's left over after perfect diversification, and unsystematic risk is what's left over after systematic risk is accounted for.
Perfect diversification consists of allocating your portfolio across all positions weighted by market capitalization. If your holdings are concentrated in a particular firm or sector beyond that, then you don't have perfect diversification, and unsystematic risk is not eliminated.
Elimination of unsystematic risk doesn't require no exposure, it requires equal exposure (all in, cap-weighted). Equal exposure still has systematic risk, but not unsystematic.
Diversification across companies and sectors doesn’t help with systematic risk, but does diversifying across asset classes (stocks, bonds, real estate, commodities) meaningfully mitigate it?
Systematic risk is often given in terms of just the stock market. So diversifying over asset classes, and different markets (i.e. foreign markets) can reduce risk below the systematic risk of the stock market. So it would arguably be more accurate if people in finance used "systematic risk" to refer to all investments available, but that is not common practice, and has there are practical barriers to working with that.
If assets become highly correlated during market crises (e.g., 2008), does diversification provide real protection?
Higher correlation tends to reduce the benefits of diversification, but as long as the correlation isn't perfect (which it never is), there is still some benefit.
Also, something to consider is that correlation is the square root of the coefficient of determination. The coefficient of determination is a measure of what percentage of the variance is attributable to common movement. That is, if most of the variance of a stock consists of the stock moving the same way as the rest of the market, then the correlation will be high. In other words, a stock's variance can be decomposed into two components. One set of terminology for these components are "explained" and "unexplained", where the "explained" component is variance that follows the market, while "unexplained" variance is variance particular to the stock. The coefficient of determination is then the ratio of the explained variance to the total variance.
So there are two ways that correlation can go up. If the amount of unexplained variance goes down, then the explained variance becomes a larger percentage of the total. Or, if the explained variance increases, that will also result in it being a larger percentage. Thus, a higher correlation doesn't necessarily mean lower particular variances. The higher correlation of 2008 was caused largely by the explained variance going up, rather than the unexplained variance going down (in fact, my guess is that the particular variance also went up, just not as much).
So while the decrease in risk achieved by diversification was lower as a percentage of total risk in 2008, that doesn't mean that the decrease in risk was lower, in an absolute sense.