Business investment accounts for around 10 per cent of GDP, but is one of the most volatile components of demand. Despite a large volume of work developing investment theories and testing them, their empirical performance is generally poor and the effectiveness of tax policy remains unclear. Chapter 1 relaxes the common assumption of capital homogeneity by estimating a Q model with multiple assets. This is done using a detailed establishment level panel. The main findings are: (a) the dataset shows clear evidence of inaction, irreversibility and heterogeneity; (b) the Q model assuming quadratic adjustment costs performs slightly better than in most previous literature; (c) the performance of the Q model is significantly improved when applied asset-by-asset; (d) asset-by-asset estimation avoids autocorrelation problems; and (e) for most asset-by-asset regressions Q is found to be a sufficient statistic. Chapter 2 investigates whether the poor empirical performance of the aggregate Q model can be explained by the exclusion of intangibles. The inclusion of intangibles improves the empirical performance of the Q model in a number of ways: (a) estimated adjustment costs are lower; (b) explanatory power is greater; (c) predictive power and parameter stability is improved; and (d) average q remains significant in a Q model with intangibles that has additional regressors but not in a standard Q model. Chapter 3 considers whether tax policy can be used to boost the long-run level of investment. The main findings are: (a) tax changes have had large impacts on the user cost of capital; (b) aggregate regressions give estimates of the user cost elasticity in the range -0.14 to -0.27; (c) tax policy can have significant impacts on the long-run level of the capital stock and investment; and (d) a natural experiment approach supports this finding by showing strong impacts of taxation following major reforms.