This thesis investigates weak identification when a forward looking Taylor rule is estimated with GMM. Lagged values of the nominal interest rate, inflation and the output gap are used as instruments. In the New Keynesian sticky price model, the strength of identification is determined by the dynamic structure. Without any backward looking terms in the inflation and output gap equations, serial correlation in inflation and the output gap shocks is necessary for identification. The order of the AR process in the shock terms determines the number of lags that are relevant instruments. Simulation of a DSGE model shows that higher persistence in the shocks increases the strength of identification. Increasing the degree of interest rate smoothing weakens identification. Monetary policy that leads to an indeterminate equilibrium increases identification, compared to a determinate equilibrium. Identification robust inference methods show that the forward looking Taylor rule is weakly identified for Norwegian data. Even though GMM estimates produce tight confidence intervals, we observe large areas of disagreement between identification robust confidence sets and confidence sets based on standard GMM inference. Possible explanations are suggested. The high degree of policy intertia observed for monetary policy will likely lead to weaker identification. Inflation expectations seem to be well anchored. This reduces the explained variation in the endogenous variables from the instruments, and hence weakens identification. Inconsistency, or discretion, in Norges Bank reaction pattern will also make identification more difficult.