The answer is 2n−2. Write p(x)=anxn+⋯+a1x+a0 and p(x)2=b2nx2n+⋯+b1x+b0. Note that b0=a02 and b2n=an2. We claim that not all of the remaining 2n−1 coefficients b1,…,b2n−1 can be negative, whence the largest possible number of negative coefficients is ≤2n−2. Indeed, suppose bi<0 for 1≤i≤2n−1. Since b1=2a0a1, we have a0=0. Assume a0>0 (or else replace p(x) by −p(x)). We claim by induction on i that ai<0 for 1≤i≤n. For i=1, this follows from 2a0a1=b1<0. If ai<0 for 1≤i≤k−1, then 2a0ak=bk−i=1∑k−1aiak−i<bk<0 and thus ak<0, completing the induction step. But now b2n−1=2an−1an>0, contradiction. It remains to show that there is a polynomial p(x) such that p(x)2 has 2n−2 negative coefficients. For example, we may take p(x)=n(xn+1)−2(xn−1+⋯+x), so that p(x)2=n2(x2n+xn+1)−2n(xn+1)(xn−1+⋯+x)+(xn−1+⋯+x)2. For i∈{1,…,n−1,n+1,…,n−1}, the coefficient of xi in p(x)2 is at most −2n (coming from the cross term) plus −2n+2 (from expanding (xn−1+⋯+x)2), and hence negative.