In an asset allocation process, correlations are particularly important if one includes 'alternative investments' such as real estate, commodities and hedge funds, which have been proclaimed to provide diversifying benefits within the overall portfolio context. While many studies have found that correlations between assets are time-dependent within each asset class, we focus on the correlations between asset classes to see if they change over time as a result of the business cycle. The technique of semi-correlation is used in order to differentiate asset class returns between up- and down-movements. We find that there are a number of assets for which correlations generally increase in down states (e.g. all types of equities). Hedge funds and balanced commodities also show substantially higher correlations to most other assets in down states. Furthermore, we find that gilts are the only asset class that becomes less correlated to most other assets during down markets, thus confirming their importance as a key diversifier in a multi-asset portfolio. Finally, once real estate indexes are adjusted to account for smoothing, we show that several correlation coefficients increase, suggesting that original time series may be overstating the benefits of diversification.